AI image and video generators have taken over the internet. Even if you don’t have access to Sora 2, chances are you’ve seen the videos reposted on social media. Nano Banana is free and now available in Google Lens, so there’s no shortage of AI images either. The creative possibilities are endless, so why are the same prompts used over and over?

A new report from Kapwing reveals the most-used prompts across millions of Midjourney creations, and it says a lot about what users want to see.

The data, pulled from over 4.9 million public prompts, shows that users lean heavily on famous artists, directors, franchises and even fast food brands to guide their generative AI results. Some of the names you’d expect. Others? Not so much.

Here’s what made the cut — and what it means for the future of AI-powered creativity.

According to Kapwing’s report, these were the most commonly used references across different categories:

Some results were predictable. For example, Star Wars remains a cultural giant, but others stood out. Alphonse Mucha, a Czech Art Nouveau painter whose work is now in the public domain, was more renowned than names like Van Gogh or Picasso.

And McDonald’s had more prompt mentions than any other brand, likely due to its iconic color scheme and global familiarity.

Why users keep name-dropping Wes Anderson and McDonald’s

In generative AI, prompts are starting points, and as this study shows, “Wes Anderson” is a popular place to begin for users who prefer a distinct aesthetic of symmetry, retro palettes and quirky set design. Similarly, “McDonald’s” evokes neon lighting, fast food signage and nostalgic Americana.

This trend helps users bypass long descriptions and rely on shared cultural context. Want a cyberpunk dystopia? Try “Blade Runner.” Dreaming up a baroque floral pattern? Use “Mucha.”

But there’s a downside to all this name-dropping.

The AI prompts killing creativity

As AI image tools rely more heavily on the same cultural references, the output starts to look... the same. It’s called “model collapse,” and it happens when AI systems keep training on each other’s content, slowly flattening originality in favor of the average.

In other words, if everyone prompts “Wes Anderson,” eventually every image starts to look like The Grand Budapest Hotel on loop.

Artists have noticed, and some aren’t happy. Greg Rutkowski, one of the most-referenced fantasy illustrators in Midjourney, is part of a class-action lawsuit alleging his style was used without permission. He’s joined by other creators trying to put guardrails around what AI can and can’t remix.

The future of AI image generation

This data confirms what many of us already feel. AI slop is flooding the internet. But there are a few positive takeaways here:

• Prompt fluency is the new creative literacy. The better you are at referencing style, era lighting and camera terms, the more likely you are to get impressive results.
• Public domain names have an edge. Artists like Mucha and classical painters may top charts simply because their work is copyright-free and safe to use.
• We may see a “prompt backlash.” As generative art becomes saturated, expect a return to more niche, abstract, or unprompted experimentation.

Final thoughts

More from Tom's Guide

• This Claude writing hack boosted my productivity — and it's completely free
• How I save money effortlessly with ChatGPT Agent: 5 clever tips to try
• This hidden ChatGPT site has 100 genius prompts — and it’s free to use

AI art models like Midjourney and Nano Banana can wow us with wild, imaginative images. But a new study reveals their so-called creativity is just based on math. And if you’re a writer, designer or musician, that’s actually a reason to breathe easier.

I’ll keep shouting this from the rooftops: AI isn’t creative. Yes, it can generate stunning images and even string together a clever story, but what it lacks is the core human understanding that makes creativity so much more than just content. The internet is overflowing with content mixed with AI slop.

In the past few weeks, Nano Banana has blown up the internet with trends and practical uses. Every time you give the model a prompt, it comes up with something fun, weird and in many cases, jaw-droppingly good. But new research shows that the creativity behind models like Nano Banana, Midjourney and other image generators, actually comes down to very simple math. In other words, what we are seeing and calling “creative” is actually nothing more than technical.

Scientists looking under the hood of our favorite image generators found that what makes these models appear “creative” has to do with the models’ structure. Specifically, two factors.:

• Locality: The AI doesn’t visualize the whole picture at once. So, when you prompt it to create “a cat wearing sunglasses and riding a skateboard,” the model works on little areas at a time, zooming in on a corner of a puzzle before pulling back to reveal the full image.
• Equivariance: A fancy word for consistency. If you move part of the input, the output shifts right along with it; like sliding a stencil across paper.

These two aspects of the model force the system to mix and remix little fragments until something new pops out. In fact, the researchers of the study even built a bare-bones version with no training data at all, and it still generated images that looked strikingly similar to what the bigger AI models produce.

A reality check for creatives

Writers, designers, musicians and pretty much any other creative might find this study obvious. It’s something we knew all along: AI is predictable, and even more than we thought.

Models will always create within boundaries such as style, genre. But unlike machines, we can decide to break rules, when to stray from a particular style and when to include a rebellious spark. That alone is still ours.

Although it very much seems like machines are catching up to us, this study is a real reminder that AI’s “imagination” will always be mechanical. It’s remixing the pieces, not inventing new ones.

As a first-of-it's-kind study, the research analyzes the why behind why AI image generators can be "creative" so we can better control, guide and evaluate the work for diversity and safety. By understanding why AI models create what they do, we have a stronger foundation into how we can push them further by controlling scale, locality and patch mixing.

That’s actually freeing, isn’t it? Sure, AI can help me sketch ideas, but the true creativity — the decision to go off-script, to zig when the rules say zag — is still ours. We are the ones in control. Imagination and creativity are something AI can’t ever take away from us.

Bottom line

More from Tom's Guide

• These 5 ChatGPT alternatives have one thing in common — and it’s not what you’d expect
• Nano Banana isn’t just for viral trends — here’s how I use it to fix and edit everyday photos
• AI billionaire’s advice to teens: Master 'vibe coding' — here’s 5 prompts to get started

When it comes to AI image generators, Midjourney has long been the gold standard for creativity, realism and versatility. But Google’s new Nano Banana model is the fast-rising contender. Lightweight, lightning quick and already making waves with its ability to crank out viral, highly-shareable images, the Gemini app now has 23 Millon total users, marking the model's undeniable popularity.

To see how the two really stack up, I ran both through the same nine prompts designed to test their strengths across realism, style, humor, text rendering and storytelling. From photorealistic portraits to Spider-Man in Times Square, here’s where each tool shines and where they stumble.

1. Photorealism

Prompt: "A hyper-realistic portrait of an elderly woman with deep wrinkles, silver hair in a bun, wearing a red scarf, lit by soft golden sunset light, ultra-detailed skin texture."

Nano Banana generated one image that captured every element of the prompt including “lit by soft golden sunset light.” The photorealistic image is strikingly realistic.

Midjourney created more photos, yet all of them failed to get the sunset light request. Although all the images were realistic-looking, they weren’t as good as Nano Banana and did not fully nail the prompt.

Winner: Nano Banana wins for its one photorealistic image that better captured every element of the prompt.

2. Fantasy worldbuilding

Prompt: "A floating castle made of glass and crystal hovering above a waterfall, with dragons circling and villagers watching from below."

Nano Banana created a breathtaking fantasy scene, with a luminous castle and cascading waterfall that felt straight out of a storybook. But the overly perfect clouds and evenly spaced birds made the image feel a little too artificial, pulling away from the magic.

Midjourney delivered four stunning images that not only captured every element of the prompt but also layered in extra details that elevated the scene beyond expectations.

Winner: Midjourney wins for creating a truly fantastical scene with elements that exceeded expectations while still carefully following the prompt.

3. Everyday realism test

Prompt: "A bustling New York City coffee shop interior on a rainy day, reflections on the floor, people working on laptops, a barista steaming milk."

Nano Banana made the rainy day evident with multiple umbrellas hanging high, which is unrealistic (they are more often next to the door and lower). The coffee shop customers are not diverse and most of the customers look the same.

Midjourney considered the diversity of NYC by adding a variety of people in the coffee shop. The viewpoint is different from what I was expecting, but it works to capture the busy coffee shop. The rain outside is less visible but it’s evident the weather is inclement based on what customers are wearing.

Winner: Midjourney wins for a more realistic image of a coffee shop in NYC despite highlighting the rain less.

4. Humor & surrealism

Prompt: "A corgi dressed as a samurai riding a skateboard through a bamboo forest, painted in ukiyo-e woodblock print style."

Nano Banana generated an image that captured both the samurai corgi and the background to better showcase the print element of the prompt.

Midjourney delivered arguably cuter images of samurai corgis. But while the imagery is clear and the dogs are in a variety of poses, the writing is not legible, which ultimately lowers the quality of the image.

Winner: Nano Banana wins for more accurately capturing the ukiyo-e woodblock print style.

5. Text rendering challenge

Prompt: "A neon diner sign at night that reads ‘OPEN 24 HOURS’ with glowing reflections on wet pavement, retro 1950s diner in the background."

Nano Banana captured the exterior of a 1950’s style diner in a photorealistic way while also spelling the words on the sign correctly. However, the puddle reflection felt overly artificial, which broke some of the immersion.

Midjourney better captured the exterior of a 1950’s style diner while also considering the inside design. While one of the images spelled “hours” incorrectly, three of the images were still more realistic than the Nano Banana’s image.

Winner: MidJourney takes this round for delivering a fuller, more believable interpretation of the prompt.

6. Cinematic landscape

Prompt: "An epic wide-angle shot of Icelandic black sand beaches with massive crashing waves, stormy skies, and a single person in a red raincoat standing at the shore."

Nano Banana generated a breathtaking image that almost felt alive. The crashing waves, moody skies and sense of depth made it feel like you were standing right there on the shoreline. It was visually striking with a story you could step into.

Midjourney produced similarly dramatic images, but the waves were so massive they bordered on overwhelming, swallowing up the scene. With no land visible in the distance, the compositions felt more like a fantasy seascape than a realistic ocean view.

Winner: Nano Banana edges ahead here for creating a more balanced, believable image that captures the beauty of nature without losing realism.

7. Abstract / stylistic

Prompt: "A cubist painting of a jazz band playing in a smoky club, exaggerated shapes, bold colors, Picasso-inspired."

Nano Banana once again proves that it’s not about quantity, but quality. The image generated, compete with frame, showcases a Picasso-inspired piece as something you could imagine hanging in a gallery.

Midjourney captured the style reasonably well, but its four outputs lacked depth and variation. Each looked nearly identical, leaving none that stood out as memorable or unique.

Winner: Nano Banana wins for originality and delivering an image that feels like genuine art.

8. Cute & viral-friendly

Prompt: "A baby elephant in a kiddie pool full of rubber ducks, wearing oversized sunglasses, pastel cartoon style."

Nano Banana delivered exactly what the prompt called for, generating an image that looked straight out of a children’s storybook or coloring book. The playful lines and bright “cartoon style” captured the whimsical feel perfectly.

Midjourney took a different approach, leaning into more realistic imagery. While the results were undeniably cute (and I kind of want to hang this series of images in my kids’ bathroom), it ultimately missed the mark. The request was for a drawing, not a near-photorealistic render.

Winner: Nano Banana wins this round for staying true to the prompt and nailing the cartoon style.

Overall winner: Nano Banana

More from Tom's Guide

• Nano Banana just broke the internet with these viral trends — I tried these 5 prompts and I'm blown away
• I tested ChatGPT-5 vs Gemini Nano Banana with 9 AI image prompts — and there's a clear winner
• I found ChatGPT-5's biggest blind spot (Memory is broken) — my workaround is shockingly simple

The hype around AI is hitting a fever pitch, but even OpenAI CEO Sam Altman, one of the leading architects of the AI boom, says things may have gone too far.

In a recent dinner with journalists, Altman compared today’s AI enthusiasm to the dot-com bubble of the late ’90s, warning that while the core technology is real, the investor frenzy may be setting the stage for disappointment.

The Verge reported Altman as saying, “Yes, investors are overexcited about AI,” adding, “Smart people get overexcited about a kernel of truth.”

What is an AI bubble?

An AI bubble refers to a surge in investor enthusiasm and startup funding that inflates valuations beyond reasonable expectations, often detached from actual revenue or real-world impact.

It’s not unlike the housing bubble of 2008 or the crypto frenzy of 2021; lots of money, lots of momentum and a growing risk of collapse if reality doesn’t match the hype.

In the context of AI, that means:

• Startups raising hundreds of millions based on vague promises of “transformative” tech.
• Companies tacking on “AI” to boost valuations without clear use cases.
• Talent wars inflating salaries to unsustainable levels.
• Investor fear of missing out (FOMO) driving reckless bets.

Why this matters now

AI tools like ChatGPT, Claude, Gemini, and Midjourney have already reshaped everything from coding and writing to education and marketing. But Altman’s warning suggests that not every AI startup will be a winner, and many may be riding a short-term wave of buzz rather than building lasting value.

And yet, despite the caution, OpenAI is still spending big. Bloomberg reported that Altman confirmed the company plans to spend trillions of dollars on infrastructure, including next-generation data centers and AI chips. Altman underscored that demand already exceeds supply in a conversation with head of TED, Chris Anderson.

Bottom line

More from Tom's Guide

• Maximize ChatGPT-5's potential with this one game-changing prompt
• I just tested ChatGPT-5 vs Deepseek with 9 prompts — and there's a clear winner
• GPT‑6 could change how we use AI forever — here's what we know

I’ve been using AI to generate images for about three years now. In that time, I’ve seen it go from a fun game that could just about replicate the theme you're asking for (emphasis on the just about here), to a movement that is near-indistinguishable from human artwork.

Nowadays, we’ve got a bunch of AI image generators to choose from. In fact, just about every company under the AI sun has one built in now. But in those early days, there were two that stood out: Dall-E (now part of ChatGPT) and Midjourney.

When Midjourney first launched, it was my favourite AI image generator. It could make exciting images and, for where AI was at the time, it could do them well. However, when I tried Midjourney’s latest upgrade, V7, it felt so behind the competition.

So what went wrong? Well, after reading a thread of Reddit users pointing out how I didn’t know how to prompt for Midjourney, I spent some more time on the AI tool. I realized that, while it is by no means the best option for most people, it might just be the most fun.

What is the problem with Midjourney images?

AI tools have gone from a niche interest to a tool that every single person you know uses. With this change, we’ve seen the models grow to understand human prompts, filling in the blanks that you might have missed.

This is an area where ChatGPT absolutely thrives, especially in images. When it comes to image generation, ChatGPT isn’t going to take a risk. It is easy to use, requiring little effort in terms of prompting, and will generate what you ask for in the style you ask for it in.

Midjourney, on the other hand, takes risks. This can be seen in its customization alone. Before you can even use it, you need to rate over 200 images, helping the model understand your own personal preference for art styles.

There are sliding scales for how weird your images are, as well as changing how stylized they will be or the variety between them. Unlike ChatGPT, Midjourney generates multiple images in one go and then lets you upscale them, make them even weirder, and try again.

I asked both models to generate cyberpunk worlds using prompts that work best for each model. Both generated accurate and genuinely interesting images for this. I then tried again, asking for a slight variation of the image.

Midjourney re-styled. While it was still accurate to the given prompt, it was a completely different look, trying something new. ChatGPT, on the other hand, got caught in a cycle of generating near-identical but highly detailed images, avoiding pushing the boat too far out from what I’d asked.

Midjourney is by no means as consistent as ChatGPT, and even with highly detailed, thought-out prompts, it can still get things wrong, smudging faces or leaving giant blurs where objects should be. However, Midjourney is also in the midst of a training period. In theory, its quality should start to improve over time as version 7 develops.

If you want to embrace the weirdness of AI images, diving headfirst into fantasy worlds, bizarre art styles, and storytelling, Midjourney can be a great place to go. Especially with the option to save images in collections, and a gallery of the best artwork Midjourney users have made, complete with their prompts for inspiration.

If you want accuracy and the ability to get consistent results from a simple prompt, ChatGPT should remain your image generator of choice. There is no learning curve here, and the experience is intuitive, even if you have absolutely no experience with AI.

More from Tom's Guide

• Google Gemini adds new image-editing tools — here's what they can do
• I use ChatGPT to plan all of my vacations — 5 essential prompts to try
• Get started with Hume AI — how to set up voices and talk naturally

Both Midjourney and ChatGPT have recently released new versions of their AI image generators. Historically, these have been two of the best options out there, pioneering the space for what has come.

But, when put against each other, which is best? Midjourney V7 or ChatGPT 4o image generation?

I put ChatGPT vs Midjourney to the test using seven different prompts to see which is the best AI image generator. These test everything from the model’s ability to understand context, recreate complex shapes and think creatively to make images.

ChatGPT vs Midjourney V7: The rules

While both models create images, it can be hard to make this a fair matchup, mostly because of the amount of settings Midjourney allows you to change. With that in mind, these were the steps I took first.

For Midjourney, I used version 7. This is the latest version but it is still in an experimental phase. I also tried the prompt with both personalization on and off (the setting that adds your preferred art styles to images).

Midjourney produces four versions of each image compared to ChatGPT’s one attempt. In all cases, I chose the best image from Midjourney and upscaled it (asked for a higher-quality version).

1. Photorealism

Prompt: Create a photorealistic image of a puffin flying over a cliff face with water below. In the background is a mountain range. It's a sunny day and below the puffin is two people looking at it through binoculars

ChatGPT

This hits almost all of the marks. The image, while potentially over-saturated, is photorealistic. There is a puffin flying over a cliff face, there is water below and there is a mountain range in the background.

On top of those points, it included the two people looking through the binoculars. Sure, they aren’t looking at the puffin but otherwise, this is pretty spot on.

Midjourney

There is a lot going on here. I can’t disagree that everything has been included. Mountains in the background, a puffin, two people with binoculars, and even water and mountains in the background.

However, let’s address the elephant (or puffin in this case) in the room. The puffin is giant and could take on Godzilla if needed. The image also isn’t really photorealistic, looking slightly more like an oil painting than anything.

Even with puffin sizing issues aside, I still think ChatGPT understood the cues more accurately. Both models created water underneath a cliff face, but ChatGPT understood the context of the prompt more accurately.

Winner: ChatGPT wins this one in just about every single way. While I'd love for puffins to be giant mythical creatures, Midjourney just misunderstood way too much context here. ChatGPT, on the other hand, nailed the brief.

2. Complicated prompts

Prompt: A large market with a stall selling fruit, one selling dresses, and one selling ceramics. In the background is a river, and in the far distance is a forest. A man hands a woman money in front of one of the stalls, and two kids are running through the middle. In the sky is a hot air balloon

ChatGPT

A lot was going on in this prompt, and it could be easy for an AI model to ignore some of it. However, all of the key details are here. The hot air balloon, the two kids running through the middle, and the man handing a woman money. It is also, clearly, a hot day, and you can see the market selling fruit, ceramics, and dresses.

Despite all the details required, ChatGPT produced a high-quality and very detailed image.

Midjourney

While Midjourney achieved the same image, it was the smaller details that were off. When zooming in, faces aren't complete, the two people's hands are morphed together, and most of the background is a blur.

Winner: ChatGPT takes this one. While both look correct at a glance, and mostly get all of the features asked for, Midjourney is just missing too many of the finer details.

3. Adapting real images

Prompt: Turn this image into a Renaissance portrait

ChatGPT

With this prompt, ChatGPT essentially turned my image into the stylings of the Mona Lisa. Again, I can’t really fault the model’s work here. It put the exact photo I supplied into the stylings of the Renaissance era.

It also does a good job of keeping all the features like headphones, the background, and the clothes I am wearing, while keeping to the theme.

Midjourney

Yes, this was the best of the four attempts Midjourney gave me. I do see where the model was trying to go here. It just couldn’t quite make it.

I even tried altering the prompt slightly to make it clear I wanted it in the style of a painting and that made things worse. I assume the brown border is also supposed to fit the theme? It's hard to tell.

Winner: ChatGPT has proved that this new model thrives when it comes to putting a creative twist on your own images, and this is more proof of that. It did exactly what I asked for. It seems like Midjourney got halfway through and gave up.

4. Movie posters

Prompt: Create an exciting poster for this film: A Cyberpunk movie set in the year 2250. It is set in a big bustling city. The film is about a detective set back in time to stop an upcoming war from happening

ChatGPT

It’s not the most exciting poster ever, but ChatGPT definitely nailed the brief here. Our detective takes centre stage, with a bustling (and rather futuristic city) nestled in the background. It did take the prompt quite literally for text, adding the requested data with a slogan.

Overall, it’s impressive. The detective is detailed with a neon light shadow on his back, there’s a flying car in the sky, and, while slightly crudely drawn, lots of futuristic skyscrapers.

Midjourney

What Midjourney lacks in detail here, it makes up for with style. Arguably, the skyscrapers look better here, and there is a lot more to see in this image.

Sadly, Midjourney falls behind with its blurry details. The images on the ground have morphed, the car and motorbike have glitched, and there are lots of bizarre details in the background.

While it’s more interesting, there is just too much wrong here.

Winner: ChatGPT did everything I asked for and made a poster that I could put out into the world and no one would bat an eyelid (other than the incredibly boring film title).

Midjourney, on the other hand, just got too many things wrong here. I do, however, like the direction it was going in.

5. Text generation

Prompt: Make an image of a poster, on the poster it says: "The band AI image generator - playing here tonight at 8pm! Covers of all your favorite hits" Stylise the image as if this is a poster for a band playing at a popular venue

ChatGPT

There’s a bit of a theme with ChatGPT’s image generation. Detail often trumps style. This poster did everything I asked for, and more importantly, got all of the text exactly right.

ChatGPT, just a few months ago would have struggled with this, so it is exciting to see how far it has come.

While the poster is boring, it has hit the brief and achieved a tricky challenge for AI models.

Midjourney

I appreciate that Midjourney made this more of a poster at a venue, putting it on the wall outside. I also like the energy it was going for with the picture of the band in the middle.

However, other than the words “The band” not a single bit of the text is readable. Compared to ChatGPT’s ability to get all of the text in its entirety, this feels like a bit of a letdown.

Winner: ChatGPT might not have been incredibly interesting here, but it completed the task perfectly. As Midjourney showed, it is not always easy for AI models to deal with text in images.

6. Hands

Prompt: Make an image of a person's hands, the left one is holding an orange and the other is holding a glass of water

ChatGPT

Ah, how far AI has come. When AI image generation first came about, one of the easiest ways to identify it was hands. They would have incredibly long fingers, or fingers sticking out of the wrong places.

Now, while the hands here don’t quite look completely human, the accuracy is really impressive. Both hands have the correct number of fingers (good start), the water in the glass properly morphs the visual of the hand behind and you can see veins and nails.

Midjourney

Midjourney did a fantastic job here. What I think is especially impressive about this image is the detail. The hair on the arms, the veins, the bruising on the knuckles, and the stretch marks on the hands.

While the ChatGPT image is instantly recognisable as AI, this could pass for someone’s hands. The only noticeable issue is the finger behind the glass not looking quite right. It is also a very strange way to hold an orange, but each to their own.

Winner: Midjourney stole a win on arguably one of the best-known flaws of AI. This goes to show how far it has come. This isn't to say ChatGPT did badly, it just didn't quite match up.

7. Food

Prompt: Make a picture of a bowl of seafood pasta that would be used on a food Instagram

ChatGPT

This is the kind of food image that I would see in a cookbook and not question for a second. Even though this doesn’t exist, I want to eat it. Can’t really fault the AI model here; it did everything that it was asked for, even if the random bit of herb in the bottom is very out of place.

Midjourney

Just like ChatGPT, Midjourney did an excellent job here. This looks like a real bowl of pasta that you would get in a nice restaurant. There are even some random tomatoes and garlic scattered around, I assume for decoration.

Winner: ChatGPT takes the win here thanks to an ever-so-slightly better image quality, but like the hands, this was close.

Verdict: ChatGPT wins

Sadly for Midjourney, this wasn't even close. Of course, it is important to note that Midjourney has just released this version and it is still in an experimental stage. However, this latest version of GPT image generation is only a week or two older.

While the model's were occasionally evenly matched, ChatGPT just so often excelled where Midjourney didn't.

I do hope Midjourney sees improvements through the testing phases of version 7 as it could be such a great AI image generator.

More from Tom's Guide

• You can now use Google's AI to make videos from text — and I'm already obsessed
• I used Grok's free new Studio AI tool to make a website, research paper and browser game — here's what happened
• I'm a personal trainer — ChatGPT built me a 15-minute mobility routine for looser hips, and I’m surprised by the results

Along a long hiatus, Midjourney is back with its latest update. Version 7 brings about big changes to image quality and contextual understanding that drastically improve the images you can create.

This newest version also includes a bunch of new features, such as Draft Mode. This is the AI image equivalent of a quick doodle, getting ideas down without powering through your credits or taking too long.

But, it isn't immediately clear how to try out Midjourney V7 — there are a few steps you need to complete first.

I've broken down everything you need to know below.

More from Tom's Guide

• These 9 ChatGPT prompts helped me advocate for my child at school — #3 changed everything for me
• ChatGPT Plus is now free for students — here’s how to get it
• Google Search just got a huge AI upgrade — here’s what's new

In 1985, Back to the Future introduced audiences to a vision of 2015 filled with flying cars, hoverboards, and self-lacing shoes. Now, in 2025, while some of those predictions remain unfulfilled, the advancements in AI have brought us technologies that are equally, if not more, astonishing.

For the 40th Anniversary, the movie is heading back to the theaters tomorrow, October 31, 2025. So if you're planning to head to the theater, why not take a journey through the film's futuristic predictions and see how today's AI innovations compare.

In Back to the Future Part II, the depiction of 2015 included several technological marvels that captured our imaginations including:

• Flying cars: Vehicles soaring above traffic.
• Hoverboards: Skateboards that levitate above the ground.
• Self-lacing shoes: Footwear that tightens automatically.
• Weather control: Manipulating weather patterns at will.
• Video calls & wearable tech: Seamless communication through screens and wearable devices.
• Talking garbage cans: Waste bins that interact with users.
• AI waiters & screens: Virtual personalities serving customers.
• Smart homes: Residences equipped with automated systems.
• Auto-dog walkers: Machines taking pets for strolls.
• Robotic gas stations: Automated fueling stations.

While some of the film's predictions remain aspirational, AI has ushered in a host of innovations that echo, and sometimes surpass, those cinematic forecasts.

What they predicted (sort of) and we now have thanks to AI

Smart assistants

While I’m disappointed that we still don't have floating AI heads like Max Headroom, we do have voice-activated assistants that have become an integral part of our daily lives.

Chatbots and AI assistants like ChatGPT, Siri, and Alexa+ answer queries with human-like conversations, control smart home devices, and assist in our communications.

Wearable tech & video calls

Marty McFly's video calls offered a glimpse into a future where communication transcends physical boundaries. Today, that vision is realized through ubiquitous virtual meetings and FaceTime chats and even taking calls from our smartwatches, enabling instant interactions across the globe at any time.

Beyond video calls, wearable technology has seamlessly integrated into our daily lives. Smartwatches, for instance, have evolved into sophisticated devices capable of monitoring health metrics like heart rate, sleep patterns, and even detecting conditions such as sleep apnea. These devices keep us connected, allowing for notifications, calls, and messages directly from our wrists.

AR and AI glasses represent another leap forward. Companies like Meta have collaborated with brands such as Ray-Ban to produce smart glasses that not only function as stylish eyewear but also incorporate features like real-time language translation, weather updates, and voice assistance.

These advancements blur the lines between digital information and the physical world, offering users an immersive experience that was once the realm of science fiction.

Predictive AI for shopping & entertainment

Platforms like Netflix, Amazon, and Spotify utilize sophisticated recommendation engines powered by AI to suggest content and products tailored to individual preferences, often anticipating what we want before we can even articulate them.

AI writing & content creation

AI has significantly changed content creation. From generating images to generating a paragraph, we now have tools that assist us across a spectrum of creative and professional tasks.

These tools are adept at mimicking various writing and artistic styles, making them valuable for crafting content that resonates with practically any audience.

Smart scheduling

Modern AI scheduling assistants go beyond basic scheduling functionalities. These tools analyze your workload, priorities, and even energy levels to automatically arrange tasks and meetings in a manner that optimizes your day.

For instance, Gemini AI integrates with Gmail and Google Calendar to auto-schedule meetings and events as does ChatGPT Tasks, for an adaptive plan that evolves with your changing priorities.

AI voice cloning

While Marty McFly saw double, advancements in AI have led to the creation of deepfake technology, too. Capable of producing highly realistic but fabricated audio and visual content, innovations from Synthesia and ElevenLabs enable the creation of highly realistic yet fabricated audio and visual content.

AI voice cloning involves using machine learning algorithms to replicate a person's voice with high fidelity. By analyzing audio recordings, these systems can generate speech that mimics the tone, pitch, and nuances of the original speaker.

Similarly, deepfakes refer to synthetic media where a person's likeness is digitally altered or replaced in images or videos, often making it appear as though they are saying or doing something they never did.

Stuff we have now that they didn’t predict

Generative AI

We now have tools such as ChatGPT, Midjourney, Suno, Pika and more that can effortlessly create art, music, and even code, pushing the boundaries of creativity and automation.

Real-time translation

From smart apps like Translate Assist on the Galaxy S25 Ultra and Ray-Ban Meta glasses to innovative earbuds, today’s devices can deliver real-time translation with surprising accuracy — seamlessly breaking down language barriers and making global communication more accessible than ever before.

Facial recognition & surveillance tech

AI-powered facial recognition is everywhere — used in everything from airport security and targeted ads to the lock screen on your phone. It’s convenient, sure, but it’s also raising serious questions about privacy, consent, and where we draw the line when our faces become data.

Stuff they predicted that’s still a bust

• Self-lacing ahoes (kind of) Inspired by the film, Nike developed the Nike MAG, a limited-edition shoe featuring self-lacing technology. While not widespread, it showcases how cinematic ideas can influence real-world innovation. 
• Flying cars Despite numerous prototypes, mass adoption of flying cars remains elusive, primarily due to regulatory and safety challenges.
• Hoverboards While hoverboard-like devices exist (I have one in my garage), true levitating boards as depicted in the film are not yet a consumer reality.
• Weather control Manipulating weather patterns remains beyond our technological capabilities, with ethical and environmental concerns posing significant hurdles.
• Instant food hydrators We're still reliant on conventional cooking methods, with instant food hydration remaining a sci-fi concept.
• Fully robotic homes While smart home devices offer automation, fully autonomous robotic homes are not yet commonplace.

Final thoughts and a glimpse into 2065

Reflecting on the 40 years since Back to the Future first captivated audiences, it's evident that while some of Doc Brown's envisioned technologies remain elusive, the rapid advancements in AI have profoundly reshaped our world. As we look ahead to the next four decades, the trajectory of AI suggests a future replete with both unprecedented opportunities and formidable challenges.

By 2065, AI is anticipated to be deeply integrated into every facet of human life, driving innovations that could redefine our existence. Experts predict that AI will not only augment but, in some cases, surpass human capabilities, leading to significant transformations across various sectors.

While the specifics of the future remain uncertain, it is clear that AI will play a central role in shaping the world of 2065. The challenge lies in navigating this evolution thoughtfully, ensuring that technological progress aligns with human values and contributes to the betterment of society.

More from Tom's Guide

• ChatGPT-5 is coming 'soon' — here's what we know
• I caught my son using ChatGPT — and it made me rethink everything about parenting in the AI age
• I told ChatGPT something that I still regret — here's 7 things you should never share

Midjourney, one of the first AI image generators to emerge, has been silent for a while. But now, after over a year, they have released their latest update — Midjourney V7.

Following rapidly in the footsteps of ChatGPT’s latest image model, which has gone viral for its ability to turn images into the style of Studio Ghibli, this newest version of Midjourney promises to be its most advanced yet.

This latest version is designed to better understand text prompts and produce images that are noticeably higher in quality. This, as Midjourney CEO David Holz explained on the company Discord, can be seen in textures, bodies, and hands, as well as in objects.

Compared to previous models of Midjourney (as well as other older AI image generators), this will mean levels of quality that are hard to identify as AI. In the past, Midjourney struggled to replicate hands or complicated shapes.

Midjourney V7: What's new?

V7 can be used in two different versions: Turbo (more expensive, but quicker) and Relax, the slower but more affordable option.

There is also a new feature called Draft Mode that renders images at 10x the speed and half the cost of the standard mode. Draft images are, however, lower quality than standard but can be enhanced if you’re happy with what is being produced.

For now, a few key features will be missing from this version. Image upscaling and retexturing are notable missing pieces and won’t arrive for another few months.

Like ChatGPT’s latest image update, there are likely to be some problems in the early stages and Midjourney could introduce or remove features to deal with an increase in traffic.

However, the team behind Midjourney has noted that to use it you’ll have to rate around 200 images to build a personalization of Midjourney. This tunes the model to your own preferences. While this has been possible on previous Midjourney models, this is the first time it is switched on by default.

You can sign up for Midjourney via their website. V7, once you’ve personalized your profile through the step above, is available either via Discord or the Midjourney website.

More from Tom's Guide

• How to get better results from ChatGPT by asking one simple question
• I just tested NotebookLM's new features — and this AI tool blows me away
• OpenAI's smartest model could cost up to $30,000 per task according to estimates

DeepSeek is the hot new player in the AI scene. The recent release of its open-source DeepSeek R1 AI model, and the sudden surge in popularity, could potentially be the biggest AI event since the launch of ChatGPT back in 2022. But DeepSeek isn’t stopping there, as it has just launched upgraded image models — named Janus Pro.

The Janus Pro models have been launched on GitHub and Hugging Face, available in one-billion and seven-billion parameter models. According to DeepSeek, the latter is able to compete with both Stable Diffusion and Dall-E 3 in benchmarking tests, though this hasn’t been thoroughly tested by third pirates right now.

The big thing about DeepSeek is that its models are apparently being trained on a fraction of the budget of western AI models, and with older AI chips. If this is true, and Janus Pro benchmarking holds up to scrutiny, then this is a pretty big deal and could be a problem for the best AI image generators. It’s no surprise that Silicon Valley seems to be worried.

Janus Pro’s image uploads are restricted to 384 x 384 pixels right now, though the actual output limitations aren’t clear. Per Android Authority, we know that the Hugging Face demo can produce 768 x 768 pixel images. Still, if DeepSeek’s example images are anything to go by, Janus Pro 7B is a significant upgrade compared to the original Janus model.

We haven’t had a chance to test the model for ourselves, so we can’t comment on the model’s capabilities compared to other image generators. But, considering how powerful DeepSeek’s chatbot has already performed compared to the likes of ChatGPT, we’re hoping for good things.

Of course, the AI isn’t without its issues, with widespread outages affecting the DeepSeek chatbot’s performance. Apparently, “large-scale malicious attacks” are targeting the platform, to the point where DeepSeek has now limited user registration to people with China-based phone numbers. However this measure is supposed to be temporary, and email registrations seemingly recommenced shortly afterwards.

You can find out more about DeepSeek in our in-depth explainer, and see how its chatbot compares to ChatGPT in our head-to-head test.

More from Tom's Guide

• Leaked memo reveals Apple’s AI plans for 2025 — this is what the company is focusing on
• I write for a living — and this AI transcription software is a true game changer
• I test AI chatbots for a living and these are the best ChatGPT alternatives

Midjourney is one of the original and most important AI image platforms around. Continually updating and now with more than 21 million users, it is finally branching out into new approaches to content creation.

Patchwork is an experimental new "infinite canvas" that you can use to create and collaborate on AI image content in a way that maintains consistent style and character. It uses Midjourney's technology but is currently independent of the app.

The new project is led by Max Kreminski, head of the Midjourney Storytelling Lab, and is being presented as a "world-building platform." Midjourney has touted this concept for some time, and it plans to evolve into 3D worlds in the future.

I'm not entirely convinced of the need after briefly trying it out. This could be because it wasn't fully working or because it is still clearly a very early beta release — but it feels more like a gimmick than the future of AI content creation.

What is Midjourney Patchwork for?

At first glance, Midjourney Patchwork looks a bit like those "infinite webpages" that were popular 15-20 years ago where you could place your advert or banner assuming you could find space.

When you first open the canvas, you're presented with a starting block and a menu bar with items like Character, Portal, and Personalization. Tapping one lets you add that type of content to the canvas.

You can create characters, places, and more within the canvas, linking them together as a unified world with a consistent style. This can also pull in your own personalized style from Midjourney if you've created one.

You create a new world by adding a Portal to the canvas. This triggers an initialization screen where you can outline world styles and a prompt to initialize the concept of that world. It then produces a new canvas with some simple assets linked to your choices in setup.

The theory is that when you create new images within that space, they will draw on those styles and reference other characters to build consistency.

Kreminski says this is a research preview and requires a Google Account to be linked to an existing Discord-based Midjourney profile. I think there is potential for this to be useful — sort of a cross between a designer's mood board and sketch pad but for AI. However, in its current implementation, it is a little on the clunky side.

More from Tom's Guide

• I tried Runway Gen-3 Turbo and it let me create realistic AI videos in seconds
• 5 best AI tools I use every day — tried and tested
• Gemini's free version beat Gemini Advanced in our tests — here's how

One of my favorite features of Midjourney is its ability to apply different styles to the same prompt and completely change the way they look. Style References can be numeric codes, a link to another image or the word ‘random’ added after --sref.

Midjourney stands firm as the most high-profile artificial intelligence image generator despite increasing competition from the likes of Ideogram, Flux and Adobe Firefly. It still ranks at the top of my best AI image generator guide and style references is among the reasons why.

There could be billions of Style Reference codes you can apply to any prompt in Midjourney as it is any number or combination of numbers from 1 digit to 10 digits. Some libraries have been launched including srefcodes and midjourneysref with several thousand examples.

I’ve tried to pull together a selection of seven Midjourney style reference codes I find particularly useful and engaging, and applied them to the same prompt to show how much of a difference a style reference can make to a final AI image.

Trying out style references

I’ve tried to come up with an image prompt that will work across disparate styles which could dramatically change the color and lighting of the resulting generation.

The prompt: “A solitary lighthouse on a rocky coastal cliff at golden hour. Waves crash against the rocks below, creating sea spray. Wild coastal flowers grow in the foreground, swaying in the wind. Wispy clouds streak across the warm sunset sky, casting long shadows. The lighthouse is white with red trim, standing tall against the dramatic landscape. A few seabirds soar in the distance.”

To get the base image, with no style reference applied (above) I set the mode to raw, set stylization to 0 and turned off personalization. I then asked Claude to define the style from each image after it was generated.

1. A pastel scene (--sref 1324941598)

The first style reference was one that generated a dreamlike feel with a pink hue. To use it just add --sref 1324941598 to the end of your prompt.

Claude described it as looking "romantic" with an "ethereal quality with delicate cherry blossom-like flowers in the foreground."

2. The old photograph (--sref 1435990713)

For the second style reference we're going back in time with a sepia toned effect making our prompt appear more like an old photograph. To use it simply apply --sref 1435990713 after the image prompt.

Claude described it as having a "cinematic quality with enhanced contrast and deeper shadows." Of all the images its the one I want to use with Runway to turn into a video.

3. Purple wildflowers (--sref 1798457390)

If you like more dark content this style reference is more for you, capturing a purple and ethereal aesthetic. Simply use --sref 1798457390 for this look.

Claude said of the image: "The lighting is more subdued and moody, creating a somewhat gothic or melancholic atmosphere. It feels like a scene from a dark fantasy game or film." It is also the only one to add a person.

4. Black and White (--sref 1872206420)

This is the only black and white style. It also made the lighthouse more box-like and made the image over exposed. Use --sref 1872206420 to get this style.

Where I assumed over exposed, Claude was more polite suggesting that "The image is stripped down to its essential elements, creating a graphic, almost abstract quality."

5. Too much yellow (--sref 2579690665)

This next style is full of color and is my favorite of the styles used here. It has a digital art feeling to it and includes my favorite color — yellow. To use it type --sref 2579690665 after your prompt.

Claude kept it simple, stating that "This version feels closest to traditional landscape photography with enhanced color processing."

6. Yet more pink (--sref 20240916)

The penultimate style in this list is the second one with a pink hue but looks remarkably different to the first. The sky feels otherworldly and gives the air of a storm coming soon. Use --sref 20240916 to get this color.

Claude described it as a Kawaii aesthetic, describing the lighthouse as clean with graphic lines offset against a dreamy background. It says it is "like something you might find in a lo-fi animation or peaceful gaming environment."

7. Dark and stormy night (--sref 448502317)

Finally, we have a style that left our image barely looking like a photograph, more like something that has been etched or sketched. To use this type --sref 448502317 after the prompt.

Claude described it as being "reminiscent of old maritime engravings or illustrations but with a more contemporary, high-contrast approach."

More from Tom's Guide

• I tried Runway Gen-3 Turbo and it let me create realistic AI videos in seconds
• 5 best AI tools I use every day — tried and tested
• Gemini's free version beat Gemini Advanced in our tests — here's how

Midjourney has announced that an expansion to its image editor is rolling out to a select group of users. The expansion allows users to edit external images within Midjourney for the first time, albeit with more restrictive moderation and privacy protections in place.

This moves Midjourney's tool beyond just being able to edit generative AI images and puts it further in contention alongside the best photo editing software like Photoshop or Affinity.

This is ironic, because the likes of Adobe have been busy introducing a bunch of AI tools into its legacy software like Photoshop or Lightroom. In addition to the ability to edit external images, Midjourney is also now making it possible to re-texture materials, surfacing and lighting within the editor.

How does this works in practice? For example, you could upload an image of a living room and then have Midjourney add or remove items, change the texture or color of the walls and even adjust the lighting to make it seem like the image was taken at night.

Midjourney has made the caveat that external image edits will only be available to those with yearly memberships, members who have been subscribers for the past 12 months, and members with at least 10,000 images. The feature is also only available on Midjourney 6.1.

"This is very new, so we want to give the community, our human moderators our next-gen AI moderation systems and heck our server clusters time to ease into everything gently," the company wrote on its Discord community and on X.

Midjourney's expansion into external images isn't just limited to pictures. Users can upload sketches or hand drawings and then use the AI to turn them into fully-fledged works of art.

And although professional graphic designers and artists may still want to use established tools for fine-tuned edits, Midjourney's effort does cover most of the basics. We've already been playing around with the editor at Tom's Guide and it's surprisingly adept at understanding which regions to change when you start using prompts.

As we wrote previously, any new additions fit seamlessly into the style and form of the existing image. In addition, the new UI is plenty responsive, which makes the overall image generation process feel more like a Photoshop-style app.

If you've been curious about getting to grips with this latest AI tool in its earlier form, here's our guide on how to use Midjourney's new AI image editor and a further guide on how to get more out of Midjourney — 5 tips that actually work.

More from Tom's Guide

• I tried Runway Gen-3 Turbo and it let me create realistic AI videos in seconds
• 5 best AI tools I use every day — tried and tested
• Gemini's free version beat Gemini Advanced in our tests — here's how

Midjourney was one of the original artificial image generation platforms, and it remains one of the best. It allows for the creation of a wide variety of images and styles, with parameters that offer incredible flexibility and control over the final look of your image.

With its move to the web earlier this year — putting an end to the reign of the Discord bot — Midjourney has become even more powerful. This includes the addition of an editor, which lets you resize the generated image and have the AI automatically fill in the gaps.

Like any powerful platform that has grown and evolved, Midjourney has its quirks. This includes hidden settings that can make a big difference, and, of course, prompting techniques that let you get the image you actually want. It compares very favorably to other AI models, often outperforming them on most metrics.

I have tried to pull together a selection of the best Midjourney tips and techniques I've picked up over the years, primarily focused on prompting rather than features — so this will still work if you're working within the Discord bot and haven't switched to the web yet.

Prompting Midjourney

Prompting Midjourney is similar to any other AI image generator: it likes you to be specific, you can use natural language and if you're vague, it'll give you its best guess.

Now, I don't claim to be the greatest Midjourney user — not even a particularly expert one. But I do use the platform a lot and have generated a considerable number of images.

Some of the creators I've seen on social media produce mind-blowing content that truly showcases the artistry of AI. I'm talking about work that either wouldn't be possible with other forms of art, or that use AI techniques to blend ideas and styles not commonly seen in other areas.

I've gathered some of the best advice I've received, along with techniques I've developed myself, to help you improve your Midjourney skills. Feel free to share any tips I've missed or ideas you think might help improve my own images in the comments!

1. Be as specific as you can

Specificity is important. Midjourney, Flux and Ideogram thrive on description and detail. The less the model has to fill in the gaps, the better the image will reflect your idea. Midjourney does do well with simple prompts — you can even give it an emoji — but detail drives direction.

Not all descriptive terms are created equal. You should focus on the primary subject, background, color scheme and elements you want to ensure are included. For example if you want a red car, specify the color.

Here is a typical AI image of a future skyline. I wanted to include a river, so made a point of including that and the bridge in my prompt. I specified not just colors but light styles and outfits.

You should generally avoid vague terms. I'd steer clear of beautiful and cool unless you specifically need those terms in the context such as a cool whip logo.

Example prompt: "A futuristic city skyline at dusk, with towering glass skyscrapers reflecting the orange and purple hues of the sunset. A river winds through the centre, with a suspension bridge lit up by soft blue LED lights. People in cyberpunk-style outfits walk along the streets, with neon signs glowing in the background."

2. Define the camera and lens type

If you want an image that is as close to a real photograph as possible, define the camera and the lens settings. Like in photography, setting the lens, angle, type and focus can affect the final output.

Use terms like wide-angle, macro or telephoto as well as whether the shot is close or wide. If it's a smartphone look you want, it helps to specify the smartphone brand and model, as Midjourney can adapt to that. It can help to specify the era too, such as '90s or 6'0s. Reference camera terms such as "50mm lens" or "ISO 100" to get a more accurate and focused image.

Experiment by changing the camera in the same prompt. In my example for this tip, I set it as a 50mm lens. Try it for yourself but make it a 35mm prime lens to put more of the scene in focus than you'd get with 50mm.

Example prompt: "A portrait of a lion in the Serengeti, shot with a 50mm lens at golden hour. The focus is sharp on the lion’s eyes, with a shallow depth of field that blurs the tall grasses in the background. The light highlights the lion’s mane, creating a warm, glowing effect."

3. Get emotional

While generic terms like cool and beautiful aren't a great idea, getting more specific with emotive terms can add a better overall feel to your image. For example, using queues like melancholic, joyful or even ominous can impact the way the vibe and tone.

As well as the specific words you can pair them with environmental details to build an atmosphere for your image. If you want something with a somber feel, you might include "dark stormy skies," or for something more upbeat, you could use "bright, sunny skies with soft, golden light," evoking a warm, cheerful mood.

If you want to experiment, why not try swapping "serene and peaceful" in the prompt below with "dark and stormy" or "bright and festive" to see how it changes the image.

Example prompt: "A serene, peaceful forest at dawn, with soft sunlight filtering through the leaves. The scene feels tranquil and calming, with a misty glow around the trees. A deer is standing in the center, its head tilted towards the light, evoking a sense of quiet wonder."

4. Iterative refinement

This is the hardest of the tips to demonstrate as it doesn't involve just a single change, and there are multiple ways to go about the process. The concept is simple though: try, try and then try again until you get exactly what you were hoping for. The idea is we're refining the image by constantly making new versions to fix little issues with the generation.

You should start with a broad, simple concept — a prompt that might be a few words long — and then build on it by adding more details with each new generation. One way to do this is to give the first image to ChatGPT and ask that chatbot to describe it. Then have it add what you want to the prompt such as a change of color, different lighting or a tree.

Midjourney also has its own describe function which can be used in the same way, or you can play with the variation and remix options built into Midjourney to adapt the image. For example, you could use the first image as a reference when creating the next version.

Example prompt (initial): "A spaceship flying through a nebula."
Refined prompt (shared): "A sleek, metallic spaceship with glowing blue engines flying through a swirling purple nebula, with stars in the background. The light from the engines reflects off the ship’s hull, casting a faint glow on the surrounding space."

5. Using parameters

Finally, the true powerhouse of Midjourney is its parameters and the control they give you over the image you are creating. These are advanced settings that can control different elements like dimensions, quality, and even creative randomness. In the Discord days, you'd need to set them with a "--" but now some of it is a button on the web.

You can define the aspect ratio, image resolution, and focal point and set a style, image reference, chaos level and creativity. There are billions of style references in Midjoruney, accessed through the --sref command with either the word random, a URL or a number.

Because I've used "chaos" in the prompt, I've included all four images generated by Midjourney from the same prompt. Chaos gives you more variety between each image.

Example prompt: "A landscape of rolling hills under a clear blue sky, with a small village in the distance. The colours should be vibrant and slightly stylised, evoking a dreamlike feel. --ar 16:9 --stylize 1000 --chaos 20."

--ar 16:9: Sets the aspect ratio to 16:9 (wide format).

--stylize 1000: Adds a strong aesthetic style to the image.

--chaos 20: Adds some variety and randomness to the generation, while keeping it controlled.

More from Tom's Guide

• I tried Runway Gen-3 Turbo and it let me create realistic AI videos in seconds
• 5 best AI tools I use every day — tried and tested
• Gemini's free version beat Gemini Advanced in our tests — here's how

Black Forest Labs released a new version of Flux last week — 1.1 Pro — claiming it outperformed all other AI image generators on most benchmarks. My early tests found its realism was impressive, speed was fast and prompt adherence exceptional.

However, for a long time, Midjourney has held the 'most realistic AI image model' crown, not just in my tests but generally. With the release of v6.1 in July it even improved skin texture on people and atmospheric rendering.

To find out whether it had finally been toppled I came up with a series of tests for both Midjourney 6.1 and Flux 1.1 Pro running through Freepik (one of our AI image platforms of the year).

Here's how our Midjourney vs Flux comparison turned out.

Creating the prompts

I created 7 conceptual ideas to put Flux 1.1 Pro and Midjourney 6.1 through their paces. I then used best practice tips for each model to craft customized prompts. This ensures each model will be able to shine as they benefit from slightly different prompting techniques. 

For example, when prompting Flux you want to focus on detailed elements such as the subject,  style, composition, lighting and color pallet. You should also include technical details such as camera settings and atmosphere to improve realism. 

For Midjourney you want to use descriptive keywords, adjectives and examples. Even more so than Flux you should focus on camera types, settings and environmental descriptions to improve realism while avoiding terms like "photorealistic" or "hyperrealistic," as Midjourney responds better to detailed visual descriptions.

For the test, the settings were largely set to default, although I switched Midjourney to RAW and turned off personalization. This was to better match Freepik's implementation of Flux 1.1 Pro.

Scoring the tests

I’ve come up with five areas to judge the two models including on detail and texture, lighting and shadows, composition and depth, color pallet and finally overall realism. You can read the full rubric including the breakdown of points for each of the five criteria in a Google Doc.

The first four of these criteria will be technical and carry the most points. They will be judged by ChatGPT using its impressive vision AI capabilities although I will have final say. The fifth is the subjective tests which will be judged by me without AI involvement.

When on default settings Freepik creates just one image using Flux 1.1 Pro, so of the four pictures created by Midjourney, I have always opted for the first image to keep it fair.

1. A mountain scene

First up we are going to see how well each model handles a landscape. Specifically a mountain at sunrise. We’re specifying the angle and want it with a high dynamic range.

Flux 1.1 Pro Prompt: A sweeping mountain landscape at sunrise, captured from a high-angle perspective using a wide-angle lens. The early morning light casts long shadows across the rugged terrain, with mist rolling over the valleys. The scene features sharp detail in the rocks, lush greenery, and clouds forming over distant peaks. Warm oranges and pinks dominate the sky, creating a dramatic and serene atmosphere. High dynamic range (HDR) captures the subtle transitions between light and shadow.

Midjourney 6.1 Prompt: Shot using a wide-angle lens, a dramatic mountain range at sunrise, the golden light spilling over the ridges, casting long, defined shadows. The camera is positioned high up, creating depth with layers of mist weaving through the valleys. The texture of the jagged rocks contrasts with the softness of the clouds and greenery below. Soft orange and pink hues fill the sky, bringing warmth to the scene. Clear focus on the details of the terrain with subtle mist movement.

ChatGPT analysis: "The Midjourney image had a more natural composition and believable depth, but both models struggled with textures that felt too smooth, leading to a slight "game-like" appearance. Midjourney scored higher due to better lighting and depth management."

I largely agreed with ChatGPT here, while both were compelling images and in some respects, Flux had some more realistic elements, overall the Midjourney image 'felt' more real.

2. A city in a blur

For the second test we’re going to create a more complex scene, moving from the mountains to the city where the AI has to depict cars and pedestrians, different lights and even water. 

Flux 1.1 Pro Prompt: A busy city street at dusk, shot with a medium focal length lens to capture the motion of cars and pedestrians. Neon signs glow brightly, reflecting off wet pavement from a recent rain. Motion blur highlights the speed of passing vehicles, while people walking on the sidewalk are crisp and detailed. The lighting creates a mix of cool blues and vibrant reds, giving the scene a fast-paced, energetic vibe. Street-level composition emphasizes the urban density, with buildings towering in the background.

Midjourney 6.1 Prompt: Captured with a 50mm lens at street level, a bustling city street at dusk, where motion blur from cars zipping by contrasts with sharply defined pedestrians walking on the wet pavement. Neon lights from signs and storefronts reflect vivid colours on the ground, creating a dynamic mix of cool blues and bright reds. The composition highlights the movement, with buildings stretching upward, showing the depth and energy of the city.

ChatGPT analysis: "Despite both images having an artificial quality, Midjourney’s lighting and reflections felt more grounded, whereas Flux’s version felt overly digital and lacked realism in texture and blur effects."

I think the AI may have been a little harsh on Flux but what really sold Midjourney was the way it handled the non-blurred parts of the image. It created a more realistic depiction of the scene.

3. Human emotion

With this test we’re going full human. Here each of the models needs to capture emotion, specifically a middle-aged man in deep sorrow, shot with an 85mm lens.

Flux 1.1 Pro Prompt: A close-up portrait of a middle-aged man, illuminated by soft, natural window light. His face is worn with deep lines, and his eyes are filled with emotion, reflecting both sorrow and contemplation. Shot with a 85mm lens to focus sharply on the subject’s facial features while softly blurring the background. The lighting is gentle, creating subtle highlights and shadows that emphasize the texture of his skin. The muted colour palette of warm browns and greys enhances the intimate and reflective mood.

Midjourney 6.1 Prompt: Shot with an 85mm lens, a close-up portrait of an older man, his weathered face illuminated by soft, diffused natural light from a nearby window. His eyes carry deep emotion, expressing a mix of sorrow and reflection. The camera focuses tightly on his facial features, with the background blurred into soft, warm browns and greys. The lighting enhances the details of his skin texture, creating a quiet, intimate atmosphere.

ChatGPT analysis: "Midjourney’s detail on skin texture and facial expressions felt more lifelike compared to Flux, which had a slightly artificial sheen to the subject. Midjourney had better lighting and emotional depth, making it the clear winner."

I have seen both Midjourney and Flux create significantly better and more realistic depictions of humans than we see here. Midjourney is better, and by some margin but I suspect if it was a younger person Flux would have won.

4. A look to the future

Every AI model loves a good futuristic scene and most of them focus on lots of neon lights. Here I want it to be more subtle. It has to capture a high-tech concept and holograms. 

Flux 1.1 Pro Prompt: A sleek, futuristic laboratory, filled with advanced holographic displays and robotic arms. The scene is captured using a 35mm lens with a shallow depth of field, focusing on a floating, transparent hologram in the centre. Cool blue and white lighting floods the room, reflecting off shiny metallic surfaces, while soft bokeh effects blur the background. The composition draws attention to the cutting-edge technology, with smooth reflections and sharp details in the illuminated interfaces.

Midjourney 6.1 Prompt: Shot with a 35mm lens at a low angle, a futuristic laboratory brimming with advanced holographic displays and robotic systems. The focus is on a floating, translucent hologram at the centre of the frame, with the background softly blurred. Cool blue and white lights reflect off sleek metallic surfaces, casting a futuristic glow across the scene. Sharp details in the holograms and the tech interfaces create a high-tech, cutting-edge feel.

ChatGPT analysis: "While both images were highly stylized, Midjourney achieved a more balanced composition and natural lighting, whereas Flux appeared flat and artificial in comparison."

This was the most subjective of the tests as both look good. For me though, the Midjourney image looks like it 'may' have been taken by some future camera that can perfectly image a hologram in a still photograph, so won on realism.

5. The fashion show

I will admit that association realism with the fashion industry might be a stretch, but with this prompt that is exactly what we’re doing. Set in a studio under intense light, the AI has to capture the shadows, reflections and of course — the beautiful gown.

Flux 1.1 Pro Prompt: A high-fashion runway show in a luxurious studio, shot with a 70-200mm telephoto lens to compress the scene and focus on the elegant model walking down the runway. Studio lights with high contrast illuminate the shimmering fabric of her gown, highlighting its intricate details and texture. The camera captures her confident stride with a soft focus on the audience in the background. The lighting is crisp, creating dramatic shadows and reflections off the polished runway floor.

Midjourney 6.1 Prompt: Captured with a 70-200mm telephoto lens, a luxury fashion runway, where a model in an intricate gown struts confidently under bright studio lights. The fabric catches the light, showcasing detailed patterns and textures. The composition focuses on the model, with a soft blur on the background audience. Sharp lighting contrasts emphasise the richness of the scene, with reflections bouncing off the polished runway.

ChatGPT analysis: "Both images exhibited a plastic-like shine, but Midjourney’s lighting and details were more polished. However, both lacked the necessary realism to feel like actual photographs."

The results here demonstrate why its worth taking a second look at the image. I liked both but was ready to give it to Flux for having the model facing forward, but then I took a look and the face is distorted and the skin looks like plastic.

6. An alien landscape

What better way to utilize the creative powers of artificial intelligence than to depict something that isn’t, couldn’t, or is unlikely to ever be seen for real? Let’s go for a deep purple planet with multiple moons but shot on a 16mm lens.

Flux 1.1 Pro Prompt: A vast, alien desert under a deep purple sky, lit by three moons. The landscape features jagged, towering rock formations and a shimmering crystal lake in the distance. Shot with a 16mm wide-angle lens to capture the expansive, surreal environment. The lighting casts long, eerie shadows across the landscape, with glowing bioluminescent plants dotting the ground. A vivid colour palette of purples, greens, and blues gives the scene a dreamlike, otherworldly atmosphere.

Midjourney 6.1 Prompt: Shot with a wide-angle 16mm lens, a vast alien landscape under a purple, star-filled sky with three moons. Jagged rock formations rise sharply, while a shimmering crystal lake glows in the distance. The lighting casts eerie shadows across the land, highlighting bioluminescent plants scattered throughout. The rich colours of purple, blue, and green give the scene an ethereal, dreamlike quality, capturing the scale and mystery of the environment.

ChatGPT analysis: "Midjourney created a more immersive scene with better depth and lighting, though it still retained a digital quality. Flux’s rendition was far less convincing, lacking the subtle interactions of light with the environment.

I'm not sure I even need to comment on this one as the difference is obvious. I actually think ChatGPT was fairer on Flux than it deserved. Its image doesn't just look like AI, it looks like early AI before realism was possible.

7. The moving vehicle

Finally, something that is a test of the real-world photographer — fast-moving cars. For this prompt we’re asking each model to use a 50mm lens to capture a vehicle speeding on a mountain road and ensure they capture background motion blur while keeping the car in focus.

Flux 1.1 Pro Prompt: A sleek, red sports car speeding along a winding mountain road at sunset, shot using a panning technique with a 50mm lens to create a motion blur in the background while keeping the car sharply in focus. The sun is low on the horizon, casting golden light across the car’s reflective surface, while the rocky cliffs and trees blur in dynamic streaks of motion. The vivid colours and precise detailing on the car’s body create a sense of speed and power.

Midjourney 6.1 Prompt: Captured with a 50mm lens using a panning technique, a bright red sports car races down a winding mountain road at sunset. The motion blur of the background contrasts with the sharp, focused details of the car, as golden light reflects off its sleek body. The rocky cliffs and blurred trees emphasise the sense of speed and movement, while the car itself remains crisp and powerful against the dynamic, vibrant backdrop.

ChatGPT analysis: "Despite both images feeling artificial due to their "too perfect" composition, Midjourney’s scene was more believable, while Flux’s image looked like a generation behind in terms of rendering quality."

Midjourney and Flux both created images that looked like they were from a game rather than a real photograph. The textures were too perfect. The difference is the Flux image looks like its from a PlayStation 3 game and Midjourney from PS5.

Winner: Midjourney

The reason I maintained an entirely 'me' judged section of the Rubric is because image appreciation, particularly around aesthetics, is largely personal. It isn't easily quantified and comes down to a sense of feeling — which AI can't offer. 

I gave my judgment to ChatGPT after having it analyze the image against the previous four sections of the rubric. I then reviewed its 'thinking' and didn't disagree with its findings, or notice any glaring errors in its analysis.

Midjourney 6.1, with RAW style, outperforms the base Flux 1.1 Pro model with no LoRAs or additional customization. Flux 1.1 Pro is still relatively new, so it is also possible prompting is more different to earlier versions of Flux than I thought, but the difference is enough that I'm happy to give the win to Midjourney.

The first thing that stood out to me was just how fast Flux 1.1 Pro is. Each image took less than a second to produce, whereas Midjourney images took several seconds but the Midjourney images are higher resolution by default.

The second thing I noticed was how different the two models are in their output. In previous and similar tests I've had similar results, even when using different prompts to get the same result — as the models improve they seem to be diverging more.

This wasn't a particularly close race between the two models. I suspect Ideogram and the latest Stable Diffusion models would have given similar results to Midjourney on some of the prompts and I need to work on Flux prompting.

What this has taught me though is the value of customizing prompts to each model to play to its strengths. This suggests that as the AI image sector evolves people might prefer to stick to a product they know, much like they do with other software.

More from Tom's Guide

• Apple is bringing iPhone Mirroring to macOS Sequoia — here’s what we know
• iOS 18 supported devices: Here are all the compatible iPhones
• Apple Intelligence unveiled — all the new AI features coming to iOS 18, iPadOS 18 and macOS Sequoia

Leading AI image generator Midjourney is working on a new feature that completely changes how we engage with generated images. Using new forms of 3D technology, you’ll be able to effectively step into the image.

Announced during the most recent Midjourney office hours on Discord it’s just one of several of new features in the works that could be launched between now and the end of this year. Midjourney CEO David Holz also said we’d see v7 before the end of the year, a video model and an upgraded image editor that lets you edit external images for the first time.

The startup has been working on 3D technology for some time as part of a wider vision of creating something akin to a Star Trek Holodeck where you can generate a world and use that world to make movies or games or just spend time.

How will the Midjourney 3D model work?

According to generative AI expert Martin Nebelong, Midjourney’s 3D model will be an entirely new approach, building on NeRF technology widely used in game development. A NeRF is a neural network able to reconstruct 3D scenes from a 2D image.

Few specific details have been released but Holz has spoken several times in the past about wanting to build a virtual world where anyone can interact and build on. During the recent Office Hours he spoke of wanting people to easily convert a Midjourney image to a 3D environment.

It would “allow camera motion within certain limits” and could have 60 frames per second rendering and a camera path system to allow video output. This is independent of the planned future video model that would likely be closer to the likes of Runway or Sora.

Other companies like Luma Labs, Adobe, and Meta have text-to-3D models but these tend to be object-based, rather than environments. Roblox is working on AI-generated environments and startup Cybever launched a waitlist for an immersive text-to-3D world model.

What else was announced by Midjourney?

According to a post on X by Alfonso Rosenberg, the main focus of the recent Midjourney office hours was on more immediate updates including around personalization. For example, for some modes, it will be on by default and users will be able to more easily refine results simply by making a selection out of four generated images.

There will also be an updated image editor that will allow external images to be edited within Midjourney although this will come with more restrictive moderation and privacy protections.

Midjourney’s video model could arrive before the end of the year but it will be better at illustrative styles than photographic ones. Finally, Version 7 of the Midjourney model is in the training phase and will be out in more than a month but less than three.

Rosenberg, who was quoting a Discord thread by JamesGriffing added that there were two hardware projects in the works, a new explore page was coming and they were testing a storytelling tool due to release this year for "world-building rather than just image creation."

More from Tom's Guide

• Best food delivery services: Grubhub vs Uber Eats vs Doordash
• The best cast iron skillets 2024: Tested and rated
• How to make images with AI using Leonardo

Civilization has been my favorite game series since I first played it on an Amiga in the early 90s. It combines addictive gameplay with a sense of history. We’ve finally seen Civ7 in action, and to mark the moment, I used AI to ‘imagine’ 7 ancient cities.

With Civ7, we have better multiplayer gameplay, a new civilization at each era change rather than playing one throughout, and improved graphics and animation. What remains is that same link to the history of humanity, starting in the age of antiquity and through to the future.

As I write about and use AI tools every day, I thought I’d combine my two passions and generate a series of images showing what an ancient city might look like if captured by a phone camera.

Civilization 7 looks like a significant change to how you play the game, with Firaxis drawing inspiration from the best of the decades-old series. The new game is available to pre-order, but Civ 6 is currently 90% off on Steam if you can't wait to play.

Creating images of 7 ancient cities

To generate the images, I first asked ChatGPT to refine my prompts, being as descriptive as possible about different elements you’d expect within each city. 

I then added a requirement for the image to look like a casual snapshot taken on a smartphone. This was after some early tests tended to lean towards drawing-style rather than photographic, even after adding the word photograph or realistic.

The final refinement was to add phrasing that discouraged the AI from simply rendering well-known buildings, such as the Colosseum in Rome, as it is today, rather than in its glory.

I have fully shared each of the prompts below so you can try them yourself. We even have a guide to using Midjourney. They will work with any AI image or video platform. The video at the top of the page was the Athens image animated using Runway.

Babylon

Babylon was one of the most legendary cities of ancient Mesopotamia and the heart of a mighty empire. The city was known for its architecture and cultural achievements. The town's focal point was the ziggurat of the Etemenanki and the lush Hanging Gardens. I've tried to capture that in the prompt.

The prompt: "A smartphone camera photo capturing the bustling heart of ancient Babylon at its peak. The image focuses on the newly built, towering ziggurat of the Etemenanki (Tower of Babel), fully intact and gleaming under the bright sunlight. The Hanging Gardens are lush and vibrant, cascading with greenery and flowers from freshly constructed terraces. In the foreground, the streets are alive with merchants, priests, and citizens, some in slight motion blur to suggest activity. The architecture is sharp and new, with no signs of wear. The photo includes natural lens flares from the sunlight, soft shadows, and a slight grain texture to give the image the look of a real-life smartphone photograph, capturing the energy and grandeur of Babylon."

Athens

Once a hub of philosophy, art, and commerce, Athens was the cradle of Western civilization and democracy. I've had the AI depict the Agora and newly constructed buildings in bright colors within the image. The focus is the bustling market.

The prompt: "A smartphone camera photo capturing a lively street scene in ancient Athens during its golden age. The image shows a bustling market in the Agora with newly constructed buildings in the background. The buildings are painted in bright, fresh colours, with no signs of wear or decay. Stalls are filled with fresh produce, pottery, and textiles, while Athenian citizens in clean, colourful togas walk through the scene, some mid-motion with a slight blur to suggest movement. The Parthenon is visible in the distance, newly built and gleaming in the sunlight. The image is bright, with natural light creating soft shadows and subtle lens flares, and a slight grain texture gives it the look of a realistic, candid smartphone photograph."

Rome

Rome was the center of the greatest empire of the age and one of the greatest empires in history. It was a city of such unparalleled architecture that some of it still stands today, including the iconic Colosseum.

The prompt: "A smartphone camera photo capturing a vibrant day in ancient Rome, focusing on the newly constructed Colosseum, which stands fully intact with all its arches and walls in perfect condition, showing no signs of decay or damage. The scene is taken from street level, with a lively crowd of Roman citizens in togas, chariots, and merchants in the foreground. The Colosseum is the centerpiece, its stone structure gleaming in the sunlight. The foreground features people in motion, with some slight blur to indicate activity, while key elements like a chariot or merchant stalls are in sharp focus. The image includes natural lens flares from the sunlight, a soft depth of field, and a subtle grain to mimic the look of a candid smartphone photograph, capturing the energy and grandeur of Rome at its peak."

Alexandria

Alexandria was founded by Alexander the Great, a center of learning and culture in the ancient world. The generated image shows the port city's peak, including the iconic Lighthouse of Alexandria.

The prompt: "A smartphone camera photo taken at the height of ancient Alexandria's prosperity, capturing the newly built Lighthouse of Alexandria standing tall and unblemished, guiding ships into the bustling port. The foreground shows a vibrant marketplace filled with scholars, merchants, and sailors, with goods displayed in vibrant colors. The Great Library and other buildings are visible, all in pristine condition with no signs of decay. The Mediterranean Sea sparkles in the sunlight, creating natural lens flares and warm reflections. The image features slight motion blur on some figures to suggest activity, and a subtle grain texture to give it a realistic, candid feel, as if taken by a smartphone in the heart of the ancient city."

Carthage

Carthage was a powerful city-state in North Africa and a major player in the Mediterranean trade, rivaling the might of Rome. In the image, we show the harbor of Carthage filled with shops and a marketplace.

The prompt: "A smartphone camera photo capturing the bustling port city of ancient Carthage at the height of its power. The image focuses on the newly constructed harbor, known as the Cothon, with its unique circular design filled with ships ready for trade and warfare. The foreground shows a busy marketplace by the docks, with merchants, sailors, and citizens in vibrant attire, trading exotic goods from across the Mediterranean. The grand buildings and temples of Carthage rise in the background, built from rich stone and adorned with intricate carvings, all in pristine condition. The sunlight reflects off the water and the city’s structures, creating natural lens flares and warm, soft shadows. A slight motion blur on some figures suggests activity, while the overall image features a subtle grain texture, giving it the look of a candid smartphone photograph capturing the energy and grandeur of Carthage."

Petra

Petra, also known as the "Rose City," was carved into the cliffs of Jordan and was the capital city of the Nabataean Kingdom. It was a key trade hub, and the image highlights the hewn rock-cut architecture set against camels and people.

The prompt: "A smartphone camera photo capturing ancient Petra at its height, focusing on the newly carved Al-Khazneh (The Treasury), fully intact and glowing in the sunlight. The foreground shows bustling activity with traders and camels moving through the streets, some in slight motion blur to suggest movement. The rock-cut architecture of Petra is sharp and newly hewn, with no signs of wear or erosion. The narrow canyon entrance, the Siq, is vibrant with life. The sunlight creates natural shadows and lens flares, while a slight grain texture gives the image a candid, smartphone-like quality, capturing the majesty and liveliness of ancient Petra."

Xi'an

Xi'an was the eastern terminus of the Silk Road and was once one of the most important cities in ancient China. Its height of power came during the Tang dynasty, and the image shows the vibrant life of the city, including walls and gates.

The prompt: "A smartphone camera photo capturing the bustling life of ancient Xi'an (Chang'an) during the Tang dynasty at its peak. The image focuses on the newly built city walls and gates, which are fully intact and in pristine condition, showing no signs of wear. The foreground features a vibrant Silk Road marketplace filled with merchants, goods, and Chinese officials, some in slight motion blur to suggest activity. The Wild Goose Pagoda and other newly constructed buildings stand tall in the background. The sunlight casts natural lens flares and warm reflections, with a slight grain texture to give the image a realistic, candid feel, as if taken by a smartphone in the heart of this thriving city."

More from Tom's Guide

• ChatGPT Plus vs Copilot Pro — which premium chatbot is better?
• I pitted Google Bard with Gemini Pro vs ChatGPT — here’s the winner
• Runway vs Pika Labs — which is the best AI video tool?

Midjourney, one of the leading artificial intelligence image generation platforms, has launched a powerful new web-based interface featuring a canvas editor and other enhancements to the user experience. I took the new tools and upgraded interface for a spin to see how they streamline the process of creating and refining AI-generated images. 

This is a huge step-up from the Discord-based image generator that returning users will be familiar with. Thanks to the upgraded UI, you now have access to a centralized dashboard for creating stellar artwork, browsing for inspiration, and even editing your own images with the power of AI.

Want to hear more about the new Midjourney? Keep reading!

 A unified web interface 

Previously, using Midjourney required joining its Discord server to interact with the AI via text commands. 

While it was good enough for a proof-of-concept, having to rely on an unfamiliar third-party platform was a deterrent for many new users. It was also far less user-friendly, without any drag-and-drop controls, menus, or sliders to simplify navigation.

With its new web interface currently in Alpha, Midjourney has introduced several new quality-of-life features to enhance the end user experience. Generating creative artwork with the power of AI is now easier than ever, with features like: 

• A unified "Create" tab that allows you to generate images from text prompts, with sliders to easily adjust image size, style, and quality.
• A library where you can browse images created by other users for inspiration. You can drag interesting images from here directly into the editor to use as references for your own creations.
• Integration of message mirroring between certain Midjourney web rooms and their corresponding Discord channels.
• Faster standard image generation with a 25% speed improvement, plus quicker job reruns.
• Expanded access, allowing any user who pays $10/month to access the web-based UI, not just power users.

 Canvas editor is the real star 

However, the highlight of the new interface is the canvas-based image editor. Clicking "Editor" on any of your AI-generated images now opens a new tool, which consolidates features like inpainting, outpainting, and image resizing into one powerful view. 

With the new AI-powered image editor, you can select specific areas of an image using a brush tool, then alter your text prompt to change just that area. 

For example, I had previously generated some artwork featuring a stylized rendition of a ghost peeking out of the forest canopy. By erasing the ghost from the image, then changing the prompt to "scary werewolf", I was able to transform it into a slightly more insidious version. You can see how the original image compares with the final version. The editor makes this type of refinement fast and easy.

 A major upgrade for AI art workflows 

Midjourney's web interface and editor are a substantial improvement over the previous Discord-based process. It puts all the key tools for iterating on AI image prompts at your fingertips.

While serious artists will still want to use external tools for fine-tuned edits, the new built-in editor covers the most important bases. It's surprisingly adept at understanding which regions to change when you alter the prompt. Any new additions fit seamlessly into the style and form of the existing image. Plus the new UI is slick and responsive, making the image generation process feel more like using a modern creative app.

If you've been curious about using AI image generation in your creative projects, Midjourney's new web tools are a great place to start. My only suggestion — I’d love to see more traditional image editing tools incorporated into the AI-powered canvas editor for some added flexibility.

More from Tom's Guide

• I tried Runway Gen-3 Turbo and it let me create realistic AI videos in seconds
• 5 best AI tools I use every day — tried and tested
• Gemini's free version beat Gemini Advanced in our tests — here's how

The world of AI-generated imagery continues to evolve at a rapid pace, pushing the boundaries of digital creativity. Recently, Midjourney got a huge update and now the AI giant has added a new feature to its toolset: photo editing. 

This addition expands Midjourney's already impressive capabilities, offering users even more control over their AI-generated and real-world images. With Midjourney the possibilities are endless, from creating photorealistic images, to stickers and visualizing home decor. The introduction of its AI image editor marks a significant step forward in digital image manipulation. 

This new feature provides users with the tools to enhance, modify, and reimagine their visual content. In this guide, we'll explore how to access and effectively use Midjourney's AI image editor, helping you make the most of this innovative tool. Let's get started! 

As AI technology continues to advance, new tools and features emerge regularly. While Grok can make images using Flux, other platforms offer unique capabilities, such as the ability to create realistic AI videos in seconds with Runway Gen-3 Turbo. And if you're curious about Gemini, the free model is superior to Advanced in many aspects. 

Runway, the leading artificial intelligence video platform has released a new “turbo” version of its Gen-3 model. This allows for the rapid creation of video from images.

Gen-3 was released earlier this month, initially as a text-to-video model but it soon added image-to-video and now, with Turbo, it made things faster.

To use Turbo simply select it from the list in the video creation tool, add any image (I’ve found Midjourney works well here) and an optional text prompt to describe the camera movement and character motion.

In my testing, Turbo was able to go from an initial prompt to a fully rendered ten-second video in just 15 seconds without any noticeable change in quality. We have near real-time AI video.

Putting Gen-3 Turbo to the test 

I’ve come up with five prompts for Midjourney to create a variety of scenes that will make the starting image for the video, and used the Runway prompt guide to come up with an associated text prompt.

All clips are ten seconds. Videos can be five or ten seconds long but there is no way to extend beyond that initial generation. You could screenshot the last frame of the generated video and use that as input for a new 10-second clip but I’ve kept it simple here.

Runway says there are more improvements to the model, control mechanisms and possibilities for real-time interactivity to come as a result of the new Turbo model.

1. The Ancient Tree

Midjourney prompt: "A massive, gnarled ancient oak tree standing alone in a misty meadow at dawn, with its twisted roots exposed and branches reaching out like arms."

Motion prompt for Runway “The camera starts at the base of the ancient oak, slowly spiraling upwards to reveal the full height of the tree against the backdrop of a misty dawn. The focus is on the intricate details of the bark, roots, and branches as the sun begins to rise."

The prompt should test the ability of both Runway and Midjourney to handle complex textures and more gradual camera movement. I think it handled both well.

2. The Village Market

Midjourney prompt: "A vibrant village market bustling with activity, featuring vendors selling colorful fruits, vegetables, and flowers, with people of all ages interacting under a bright, sunny sky."

Runway motion: "The camera moves through the lively village market, capturing the energetic interactions of people bartering and laughing. The focus shifts between vendors displaying their goods and customers browsing, emphasizing the market's vibrant atmosphere."

Here we are seeing if the AI can handle dynamic, human-centric scenes with lots of movement and interaction. It also needs to maintain the look of the image through the scene.

3. The Influencer

Midjourney image: "A young woman recording a vlog in a cozy, well-lit room filled with plants, books, and soft decor, with a ring light and camera set up in front of her."

Runway motion: "The camera follows the influencer as she moves around her cozy room, adjusting the lighting and camera, then begins recording her vlog. The shot focuses on her facial expressions and the warm, inviting atmosphere of the space."

With this prompt, we’re testing whether AI can simulate human expression and whether Runway can handle hand motions. It was good but not perfect, with an element of unreality.

4. The Train Ride

Midjourney prompt: “A scenic train ride through a mountainous landscape during the golden hour, with passengers gazing out the window at the breathtaking view."

Runway motion: “The camera starts inside the train, focusing on passengers looking out the window with golden hour lighting, then shifts to a view outside, capturing the beautiful mountainous scenery as the train glides through the landscape during the golden hour."

This test requires Runway to create a transition between one-shot, inspired by the image, and the next while trying to maintain the same feel. It was close but I think would have been better it was a text-to-video prompt without the starting image.

5. A Music Festival

Midjourney image: “A vibrant outdoor music festival at dusk, with a large crowd of people dancing, colorful lights illuminating the stage, and a band performing energetically."

Runway motion: "The camera sweeps over the energetic crowd at a music festival, capturing the lively dancing and flashing stage lights as the band performs. The focus moves from the stage to the crowd, highlighting the collective excitement and energy of the event."

Finally, we want to see how Runway handles a complex, high-energy scene that includes multiple points of motion. It did a great job although the dancers were all a little similar.

Final thoughts

Being able to rapidly generate a video is a significant change for Runway. It also points to a potential future higher resolution mode where you can upscale failed generations. 

AI video has come a very long way in just a year. We’ve reached a point where we can make a short film made up of ten-second clips and have it look almost real. Every new generation improves the image and motion realism.

Turbo simply makes this whole process faster, allowing for rapid iterations — which is useful as the ratio of usable to unusable clips is still about 5:1.

More from Tom's Guide

• Apple is bringing iPhone Mirroring to macOS Sequoia — here’s what we know
• iOS 18 supported devices: Here are all the compatible iPhones
• Apple Intelligence unveiled — all the new AI features coming to iOS 18, iPadOS 18 and macOS Sequoia

Flux is an artificial intelligence image generator released by AI startup Black Forest Labs in the past few weeks and it has quickly become one of the most powerful and popular tools of its kind, even giving market leader Midjourney a run for its money.

Unlike Midjourney, which is a closed and paid-for service only available from Midjourney itself, Flux is an open-source model available to download and run locally or on a range of platforms such as Freepik, NightCafe and Hugging Face.

To determine whether Flux has reached Midjourney levels of photorealism and accurate human depiction I’ve come up with 5 descriptive prompts and run them on both. I’m generating Flux images using ComfyUI installed through the Pinokio AI installer.

Creating the prompts

Both Midjourney and Flux benefit from a descriptive prompt. To get exactly what you want out of the model its good to describe not just the person but also the style, lighting and structure.

I’ve included each prompt below for you to try yourself and these should also work with Ideogram, DALL-E 3 in ChatGPT or other AI image platforms if you don’t have Midjourney or Flux but, except Ideogram, none reach the realism of Midjourney or Flux.

1. A chef in the kitchen

The first test combines the need to generate a complex skin texture with a dynamic environment — namely a professional kitchen. The prompt asks for a woman in her mid-50s in the middle of preparing a meal. 

It also asks for the depiction of sous chefs in the background and for the chef's name to be shown on a "spotless white double-breasted chef's jacket".

A seasoned chef in her mid-50s is captured in action in a bustling professional kitchen. Her salt-and-pepper hair is neatly tucked under a crisp white chef's hat, with a few strands escaping around her temples. Her face, marked with laugh lines, shows intense concentration as she tastes a sauce from a wooden spoon. Her eyes, a warm brown, narrow slightly as she considers the flavor. The chef is wearing a spotless white double-breasted chef's jacket with her name embroidered in blue on the breast pocket. Black and white checkered pants and slip-resistant clogs complete her professional attire. A colorful array of sauce stains on her apron tells the story of a busy service. Behind her, the kitchen is a hive of activity. Stainless steel surfaces gleam under bright overhead lights, reflecting the controlled chaos of dinner service. Sous chefs in white jackets move purposefully between stations, and steam rises from pots on industrial stoves. Plates of artfully arranged dishes wait on the pass, ready for service. In the foreground, a marble countertop is visible, strewn with fresh herbs and exotic spices. A stack of well-worn cookbooks sits nearby, hinting at the chef's dedication to her craft and continuous learning. The overall scene captures the intensity, precision, and passion of high-end culinary artistry.

Winner: Midjourney
Midjourney wins for the realism of the main character. It isn't perfect and I prefer the dynamism of the Flux image but the challenge is creating accurate humans and Midjourney is closer with better skin texture.

2. A street musician

The next prompt asks both AI image generators to show a street musician in his late 30s performing on a busy city corner lost in the moment of the music. 

Part of the prompt requires the inclusion of an appreciative passerby, coins in a guitar case and city life blurring in motion behind the main character.

A street musician in his late 30s is frozen in a moment of passionate performance on a busy city corner. His long, dark dreadlocks are caught mid-sway, some falling over his face while others dance in the air around him. His eyes are closed in deep concentration, brows slightly furrowed, as his weathered hands move deftly over the strings of an old, well-loved acoustic guitar. The musician is wearing a vibrant, hand-knitted sweater that's a patchwork of blues, greens, and purples. It hangs loosely over distressed jeans with artistic patches on the knees. On his feet are scuffed brown leather boots, tapping in rhythm with his music. Multiple colorful braided bracelets adorn his wrists, adding to his bohemian appearance. He stands on a gritty sidewalk, with a battered guitar case open at his feet. It's scattered with coins and bills from appreciative passersby, along with a few fallen autumn leaves. Behind him, city life unfolds in a blur of motion: pedestrians hurry past, yellow taxis honk in the congested street, and neon signs begin to flicker to life as dusk settles over the urban landscape. In the foreground, slightly out of focus, a child tugs on her mother's hand, trying to stop and listen to the music. The scene captures the raw energy and emotion of street performance against the backdrop of a bustling, indifferent city.

Winner: Midjourney
Midjourney wins again for the realism of the character. The texture quality of v6.1 once again puts it just ahead. It is also overall a better image in terms of structure, layout and background. 

3. The gardener

Generating images of older people can always be a struggle for AI image generators because of the more complex skin texture. Here we want a woman in her 80s caring for plants in a rooftop garden. 

The image depicts elements of the scene including climbing vines and a golden evening light with the city skyline looming large behind our gardener. 

An elderly woman in her early 80s is tenderly caring for plants in her rooftop garden, set against a backdrop of a crowded city. Her silver hair is tied back in a loose bun, with wispy strands escaping to frame her kind, deeply wrinkled face. Her blue eyes twinkle with contentment as she smiles at a ripe tomato cradled gently in her soil-stained gardening gloves. She's wearing a floral print dress in soft pastels, protected by a well-worn, earth-toned apron. Comfortable slip-on shoes and a wide-brimmed straw hat complete her gardening outfit. A pair of reading glasses hangs from a beaded chain around her neck, ready for when she needs to consult her gardening journal. The rooftop around her is transformed into a green oasis. Raised beds burst with a variety of vegetables and flowers, creating a colorful patchwork. Trellises covered in climbing vines stand tall, and terracotta pots filled with herbs line the edges. A small greenhouse is visible in one corner, its glass panels reflecting the golden evening light. In the background, the city skyline looms large - a forest of concrete and glass that stands in stark contrast to this vibrant garden. The setting sun casts a warm glow over the scene, highlighting the lush plants and the serenity on the woman's face as she finds peace in her urban Eden.

Winner: Midjourney
Once again Midjourney wins because of the texture quality. It struggled a little with the gloved fingers but it was better than Flux. That doesn't mean Flux isn't a good image but it isn't as good as Midjourney.

4. Paramedic in an emergency

For this prompt I went with something more action heavy, focusing on a paramedic in the moment of rushing to the ambulance on a rainy day. This included a description of water droplets clinging to eyelashes and reflective strips.

This was a more challenging prompt for AI image generators as it has to capture the darker environment. 'Golden hour' light is easier for AI than night and twilight.

A young paramedic in her mid-20s is captured in a moment of urgent action as she rushes out of an ambulance on a rainy night. Her short blonde hair is plastered to her forehead by the rain, and droplets cling to her eyelashes. Her blue eyes are sharp and focused, reflecting the flashing lights of the emergency vehicles. Her expression is one of determination and controlled urgency. She's wearing a dark blue uniform with reflective strips that catch the light, the jacket partially unzipped to reveal a light blue shirt underneath. A stethoscope hangs around her neck, bouncing slightly as she moves. Heavy-duty black boots splash through puddles, and a waterproof watch is visible on her wrist, its face illuminated for easy reading in the darkness. In her arms, she carries a large red medical bag, gripping it tightly as she navigates the wet pavement. Behind her, the ambulance looms large, its red and blue lights casting an eerie glow over the rain-slicked street. Her partner can be seen in the background, wheeling out a gurney from the back of the vehicle. In the foreground, blurred by the rain and motion, concerned onlookers gather under umbrellas near what appears to be a car accident scene just out of frame. The wet street reflects the emergency lights, creating a dramatic kaleidoscope of color against the dark night. The entire scene pulses with tension and the critical nature of the unfolding emergency.

Winner: Draw
I don't think either AI image generator won this round. Both have washed out and over 'plastic' face textures likely caused by the lighting issues. Midjourney does a slightly better job matching the description of the scene.

5. The retired astronaut

Finally we have a scene in a school. Here I've asked the AI models to generate a retired astronaut in his late 60s giving a presentation about space.

He is well presented in good health depicting a NASA logo. The background is well described with posters, quotes and people watching as he speaks.

A retired astronaut in his late 60s is giving an animated presentation at a science museum. His silver hair is neatly trimmed, and despite his age, he stands tall and straight, a testament to years of rigorous physical training. His blue eyes sparkle with enthusiasm as he gestures towards a large scale model of the solar system suspended from the ceiling. He's dressed in a navy blue blazer with a small, subtle NASA pin on the lapel. Underneath, he wears a light blue button-up shirt and khaki slacks. On his left wrist is a watch that looks suspiciously like the ones worn on space missions. His hands, though showing signs of age, move with the precision and control of someone used to operating in zero gravity. Around him, a diverse group of students listen with rapt attention. Some furiously scribble notes, while others have their hands half-raised, eager to ask questions. The audience is a mix of ages and backgrounds, all united by their fascination with space exploration. The walls of the presentation space are adorned with large, high-resolution photographs of galaxies, nebulae, and planets. Inspirational quotes about exploration and discovery are interspersed between the images. In one corner, a genuine space suit stands in a glass case, adding authenticity to the presenter's words. Sunlight streams through large windows, illuminating particles of dust floating in the air, reminiscent of stars in the night sky. The entire scene is bathed in a sense of wonder and possibility, as the retired astronaut bridges the gap between Earth and the cosmos for his eager audience.

Winner: Flux
I am giving this one to Flux. It won because it had skin texture and human realism on par or slightly better than Midjourney but with a much better overall image structure including more realistic background people.

Flux vs Midjourney: Which model wins

This was almost a clean sweep for Midjourney and it was mainly driven by the improvements Midjourney has made in skin texture rendering with v6.1.

I don't think it was as clear as it looks on paper though as in many images Flux had a better overall image structure and was better at backgrounds. I've also found Flux is more consistent with text rendering than Midjourney — but this test was about people and creating realistic digital humans.

What it does show is that even at the bleeding edge of AI image generation there are still tells in every image that sell it as AI generated.

More from Tom's Guide

• Best food delivery services: Grubhub vs Uber Eats vs Doordash
• The best cast iron skillets 2024: Tested and rated
• How to make images with AI using Leonardo

Midjourney released a surprise new update this week and it's a much bigger deal than its numbering would suggest. Version 6.1 would hint at a few minor changes but with it, we’ve got more natural looking people, improved text rendering, and even better upscaling.

According to the AI image lab, we should get version 6.2 in the next month or so, which suggests more regular iterative updates may be the norm going forward, rather than making everyone wait until version 7 finishes training and testing.

This approach makes sense as each major new version of a model will require more computing power, time, and data to scale beyond the current generation. Whereas fine-tuning and prompt improvements can make a big difference fairly quickly to current models.

The latest update also comes with better styling and improvements to personalization. To access it you just need to append –v 6.1 to the end of a prompt. 

Creating prompts to test Midjourney v6.1

I’ve tried to come up with prompts that test different aspects of the new features, particularly around skin and human features, text rendering and some of the style changes.

1. Holding a butterfly

The first prompt is an examination of how well Midjourney renders both the fingers and the complexities of a butterfly. This should push the texture generation.

Prompt: “Hyperrealistic close-up of a human hand holding a delicate butterfly, intricate details on fingertips and wings, soft natural lighting --v 6.1.”

2. A bright neon sign

The aim here is to test two things, the ability to render text accurately and stylishly within a complex scene, as well as the improved styles such as 8-bit and pixels.

Prompt: "“Welcome to Paradise" neon sign in a bustling cyberpunk cityscape, 8-bit retro aesthetic with enhanced pixel textures --v 6.1.”

3. A majestic lion

This was as much because its cool to see how well Midjourney produces images of a lion, but it also pushes it to render complex textures and lighting. 

Prompt: “Majestic lion with a flowing mane standing atop a rocky outcrop, distant savanna landscape, golden hour lighting, ultra-high detail fur textures --v 6.1.”

4. Lush greenery

Here, we're exploring how well Midjourney v6.1 can render complexity. Specific vivid flower details and leaf patterns while also showing a depth of field. 

Prompt: “Lush botanical garden with diverse exotic plants, focusing on intricate leaf patterns and vibrant flower details, soft depth of field --v 6.1.”

5. Ballet dancer leaping

This prompt is designed to test how well 6.1 captures fluid motion and dynamic movement. It needs to add a controlled illumination and render hands at a distance.

Prompt: “Full-body portrait of a ballet dancer in mid-leap, capturing fluid motion and muscle definition, soft studio lighting --v 6.1.”

6. Wise Eyes

This prompt puts Midjourney to the test in its promise of improved skin texture rendering, putting the face front-and-center.

Prompt: “Portrait of an elderly wise woman with deep wrinkles and expressive eyes, soft natural lighting emphasizing skin texture and facial features --v 6.1.”

7. Street Photography

Not only does v6.1 have to accurately render multiple people but it also pushes it to render a complex realistic scene. It has to capture individual details on multiple characters.

Prompt: “Candid street photography of a bustling city crosswalk, diverse crowd with varied expressions and poses, focus on individual details in a group context --v 6.1.”

Final thoughts

While some of the improvements might be subtle, there is a noteworthy improvement in texture generation and in upscaling. For each test, I ran the final image through a creative upscale and the improvement was more noticeable than with version 6.

I like that Midjourney is moving to focus on iterative updates as they build towards what I suspect will be a major change with v7, likely creating the start of a better world model.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breaktrhough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

Midjourney, the leading artificial intelligence image generation platform has dropped a surprise update to its core model. In version 6.1, skin on humans looks more natural and rendered text is also more legible.

There was some speculation the company would release v6.5 at some point this summer but it seems they’ve gone for a more iterative approach, with v6.2 due in the coming weeks.

The biggest changes are to people. Specifically how it handles the depiction of arms, legs, hands and bodies. Its texture mapping has also been upgraded to offer new skin textures.

Having briefly played with the new version I’d say 6.1 feels more like a major upgrade than its iterative numbering would suggest. The improved text rendering alone is worth the change.

What is new in Midjourney v6.1?

Overall there are changes to every aspect of the model. Subtle upgrades to each area improve image quality by reducing the number of pixel artifacts, enhancing textures and the way it handles certain styles such as 8bit and retro designs.

Midjourney says the new model is 25% faster when running a standard job and upgrades to the personalization model allow for improved nuance, surprise and accuracy over v6.

The company wrote on X: “V6.1 greatly improves image quality, coherence, text, and comes with brand-new upscaling and personalization models,” adding that “it’s smarter, faster, clearer, and more beautiful. We hope you enjoy our best model yet.”

One of the new changes is in how the upscale works, offering better image and texture quality to improve overall look and feel. There is a new –q 2 mode that takes much longer but adds more texture to further improve the realism of an image.

There are also more precise, detailed, and correct small image features that are perfect for eyes, small faces and far away hands.

The feature I’m most excited about is improved text accuracy. This is something all AI models struggle with but Midjourney says if you put words within quotations in a problem it will accurately render those words on the image.

How well does Midjourney v6.1 work?

To use Midjourney v6.1 simply add –v 6.1 to the end of your prompt. This works on the web and the Discord versions and will change the model you are using. I ran a few tests and the most obvious changes are to skin and text rendering.

For the first test, I gave it the prompt: “A poster for a movie called "Cats in Space" where the subheadline is "They are feline good" showing cats on the moon in a spacesuit.” This was detailed enough to direct the model and included text requirements.

The poster came out better than I could have expected, although only two versions had the right style and one of those had an accurate rendering of the headline and subheadline.

I then asked it to show a “Wide shot of a woman playing a public free piano in a train station.” This was vague enough that if the prompt following was off it would give something bizarre but it didn’t, I got exactly what I was hoping for although one version almost had her on the tracks.

Finally, I had Midjourney v6.1 generate an image of a woman and animated it using the impressive new Runway Gen-3 Alpha image-to-video functionality and it is one of the most real-looking AI images and videos I've created to date.

Overall I think this is a notable improvement to Midjourney, offering subtle but significant changes to areas the base model struggled with and is a great sign of what is to come in v7.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breaktrhough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

Midjourney, the AI-powered image generation tool, has taken the creative world by storm. While many are familiar with its ability to produce stunning artwork and concept designs, this versatile platform has a myriad of practical applications that often go unnoticed. From personalizing everyday items to visualizing complex ideas, Midjourney's potential extends far beyond the realm of digital art. 

You can create coloring book pages and even stickers, transforming simple ideas into tangible designs. And as the AI evolves, so does its capabilities. Improved realism and skin textures are on the way, with Midjourney v6.5 set to launch by the end of the month.

Midjourney's potential extends far beyond typical uses, with seven surprising prompts awaiting discovery. Intrigued to see how these lesser-known Midjourney tricks can elevate your projects? Lets get into it. 

T-shirt design prototype

Custom cook book illustrations

Birthday card

Personalized gift wrap

Home decor

Tattoo design visualization

Pet portraits: pop-up style

To further enhance your Midjourney creations, you can make images more personal with the personalize feature. Why not experiment with these stunning styles to get the perfect AI photo that matches your vision. And for those eagerly awaiting new developments, keep an eye out — Midjourney could be coming to Grok soon. 

Midjourney, the AI art sensation, isn't just for creating fantastical landscapes or surreal portraits. It's a versatile tool that can help you design custom stickers. With Midjourney, you can craft unique designs without ever touching a sketchpad. The possibilities are endless — you can even make a coloring book or a comic strip. As AI technology evolves at breakneck speed, V.6.5 of Midjourney could be out by the end of the month, the creative potential keeps expanding.

Getting the perfect AI photo does take some refining so don't expect instant perfection. In this guide, we'll walk you through the process of using Midjourney to create eye-catching stickers. Whether you're aiming to personalize your gear, craft unique gifts or perhaps launch a side hustle, we've got you covered.

Final thoughts 

After creating your custom stickers with Midjourney, you'll likely be impressed by the quality and uniqueness of your designs. The AI's ability to interpret prompts and generate visually striking images often surpasses expectations, resulting in stickers that are truly one of a kind.

To bring your digital creations into the physical world, consider printing them at home on sticker paper using a high-quality color printer. Alternatively, for a more professional finish, you can upload your designs to online sticker printing services. .

Getting the most of Midjourney by uploading images can significantly enhance your results, allowing you to build upon existing designs. For those looking to push boundaries and turn their AI images into videos, our guide on Runway VS Pika puts them head to head. And if you're contemplating cancelling your Midjourney subscription, we have a guide for that too. 

Midjourney's AI has revolutionized the way we approach creative projects, including the creation of personalized coloring books. The prompting possibilities feel endless, and you can use it to transform your prompts into intricate line drawings, opening up a world of opportunities for both artists and non-artists alike. By leveraging Midjourney's capabilities, you can generate unique coloring pages that span a wide range of styles and themes. 

The process is surprisingly straightforward: describe your vision, and watch as the AI brings it to life in black and white, ready for coloring. Whether you're crafting a book for personal enjoyment, as a thoughtful gift, or even for commercial purposes, Midjourney streamlines the creation process. 

Let's explore how to bring your colorful ideas to life using the power of AI and your imagination.

The results...

Final thoughts 

Harnessing AI to create a coloring book pages is an interesting way of utilizing AI for a creative project. Whether or not you've considered it before, it lets you create personalized art without traditional artistic skills. What's more, it makes for an excellent gift. And the ability to generate countless variations means you can curate a truly one-of-a-kind collection tailored to the recipient's interests. 

Taking personalization a step further, you could even incorporate images of the gift recipient, their pets, or favorite places into the coloring pages. Who doesn't want to see themselves transformed into whimsical characters within a magical world? Happy colouring! 

As you create your AI coloring book, keep in mind that Midjourney is constantly evolving. Midjourney v6.5 could be out by the end of the month, with improved realism and skin textures. The AI model's personalize feature makes it easier to generate images that match your vision. For more creative inspiration, check out these 7 stunning Midjourney styles to try for getting the perfect AI photo. 

As one of the best AI image generators, Midjourney can create incredibly realistic depictions of humans and scenes. But without specific prompting, the results can be somewhat haphazard, leaning more on the side of art than photo.

Even if you manage to refine the prompt to depict a photorealistic scene, you then have to further refine it to provide a degree of style or originality. That is unless you can find the right style parameters. These shortcodes can act as a shortcut to an impressive creation.

There are thought to be more than four billion style reference codes available, accessible by typing --sref and a number at the end of the prompt. You can also just provide the url of an image or photograph with the style you’re going for but ensure you have the rights first.

I’ve pulled together a handful of useful sref codes I’ve found that work particularly well with photorealistic-style images. A way to create compelling ‘photos’ with added flare using AI.

Creating the prompts

For each prompt I’ve created a different scene and scenario, keeping the actual style description to a minimum and instead relying on the style reference code.

I ran each prompt with and without the corresponding style reference to see how much of a difference using an sref code can have to the overall creativity of an image. You can discover several impressive 'sref' codes with reference images on Midjourneysref by underwoodxie96.

1. The city street scene (--sref 13)

First we start with this simple descriptive prompt “A lively street in a major city with people walking and cars passing by,” and add --sref 13. 

We went from a modern, well lit city scene looking like a photo taken with a relatively good camera to something out of a noir film, sepia-toned and with traditional looking buildings.

This shows just how much of a difference a style reference can have on the overall image. It completely transformed the picture despite none of the prompt changing.

2. A portrait of time (--sref 174971033)

Next, the prompt is one for more of a portrait. I used this prompt: “A close-up, detailed portrait of an elderly person with wrinkles and grey hair, capturing the wisdom in their eyes” and then regenerated appending this style reference: --sref 174971033.

The image generated with the sref code has a high contrast black and white feel, very intense and creates a more intensive sense of emotion. It does take away from the photorealism but its worth the sacrifice for the stark emotion visible.

3. A vintage car in the country (–sref 3267366411)

Next, we're moving into vehicles. The prompt I created was “a classic vintage car parked on a dirt road in the countryside, with rolling hills and a farmhouse in the background.” Then I added the sref 3267366411 for a digital art effect.

I love this prompt, although of the four images it generated in Midjourney only two of them actually looked good, so you might need to be more descriptive.

4. Wildlife safari (--sref 1707707235)

We've had a street view, an old person and an old car so now we're shifting gears and getting Midjourney to depict a safari scene. The prompt: "“A snapshot of a safari, featuring animals like elephants, zebras, and giraffes in their natural habitat.”

This was one of the most radical transformations from an sref. I used 1707707235 which completely changed the way the AI interpreted the prompt, in one image showing a woman wearing animal skins, in another an elephant car.

5. Concert Crowd (--sref 2169177422)

In another era I spent a lot of time photographing festivals and have an archive of images that look exactly like this one. The prompt: "A dynamic photo of a large crowd at an outdoor concert, with people cheering and bright stage lights.”

I used the sref 2169177422 which interestingly seems to have added a flashlight type effect to the image, putting specific people in stark relief and minimizing stage light.

6. Farmers Market (--sref 3879150397)

This was my favorite sref. I used the prompt: "“A colourful farmer’s market with various stalls selling fresh produce, flowers, and artisanal goods, bustling with activity." This gave me a colorful food market scene.

Appending the sref 3879150397 gave me a stunning image of a time gone by, depicting with hints of red throughout adding a stunning style overlay. This is an example of how sref can be used to enhance an image.

7. Family picnic (--sref 3200)

In our final prompt the sref actually did very little, adding a slight haze. The prompt: "Photograph of a family enjoying a picnic on a sunny day in a park, with a blanket spread out, children playing, and trees providing shade.” 

This gave us a fun picnic scene in a park with trees on a sunny day. Sharp greens and hints of sunlight through the trees. The sref, 3200, added a haze-like effect over everything  giving it more of classic feel.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breakthrough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

Midjourney is one of the most advanced, photorealistic artificial intelligence image generators available today and it's about to get even better when version 6.5 is released. With this new update, we’ll get improved realism and better skin textures — and it could launch this month.

According to a post on X by AI art expert Nick St. Pierre, Midjourney CEO David Holz said during the most recent “office hours” that work was progressing on the new update.

Holtz told people on the call that v6.5 will come with improved image clarity as well as improvements to specific areas such as hand quality and skin textures. As well as upgrades to the model there will be improvements to the web app and personalization.

AI artist Alie Jules, who also tuned in for the call, reported on X that the first release will likely be in alpha for a select number of users while they stress test the update, rather than broadly available. This is in line with previous releases and a common approach for AI companies.

What can we expect from Midjourney 6.5?

Each new Midjourney update brings with it certain features or changes to the model that weren't previously available in earlier versions. For example, version 6 brought with it better text rendering and significant improvements to photo-realism.

I suspect with 6.5 we'll see more of that with improvements to specific pixel quality and density as well as updates to the way the model renders skin and facial tones. This would allow for the creation of even more realistic-looking humans. Something Midjourney already does better than any other model I’ve tried.

In earlier office hours, Holtz suggested the next iterative update could be split into two, focusing on pixel quality in one and then hand and body coherence in the next, but it appears they are now working on a single 6.5 release.

Other updates likely in 6.5 are improvements to text rendering. While this got better with v6 it isn’t as accurate as other models like Ideogram, Phoenix from Leonardo or even DALL-E 3.

What else came out during office hours?

As well as updates to the progress on 6.5, Holtz also confirmed that they are still working on version 7, currently focusing more on the speed of generation and a completely clean slate for the personalization feature.

Personalization was introduced earlier this year to allow users to customize the way the model generates images based ont heir own preferences.

Holtz also confirmed they haven’t started the final training run for v7 and that it will need to go through testing and other tasks before release — so there’s no date available yet.

Other updates include the fact that they're running behind on building a style exploration tool but they may work on a simpler solution just to give people an idea of the many many styles available within Midjourney more quickly.

The update also confirmed 3D and video models are still in development but they may take some time, particularly video, as Midjourney is taking a more detailed approach that allows for better control and realism than current transformer diffusion models.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breaktrhough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

You can use AI image generators to boost your productivity at work or get your creative juices flowing when coming up with design concepts but have you ever tried using them to fantasize about what the perfect meal could look like?

Asking your friend group to pick the perfect meal will surely kick off a heated discussion. Tucking into a juicy burger at the end of a long week can hit all the right spots. Equally, a more sophisticated meal at a high-end restaurant with nice company can also be a well-deserved treat.

Suffice to say, we’re likely to disagree on what constitutes the perfect meal but we can compromise on what constitutes an aesthetically pleasing dish. If AI is truly going to become a more integral part of our lives it’s going to have to understand what makes us tick. 

With food being such an integral part of our family lives and culture, I was curious to know what AI image generators think of when we ask them to create the perfect meal. 

To find out, I picked 5 iconic dishes from each of French, Indian, Japanese, Italian, and Mexican cuisine and asked 5 leading AI image generators to present them with a modern twist.

1. Beef Bourguignon by Midjourney

This French beef stew featuring tender beef chunks enveloped in a lovely red wine sauce is a classic French dish coming from France’s Burgundy region. 

Midjourney seems to have slow-cooked the meat to perfection and paired it with a couple of potatoes to complete the hearty meal. The AI image generator also took an extra moment to decorate the plate.

2. Chicken biryani by Leonardo

For this round I turned to Leonardo’s Phoenix model, its new advanced standard-setting model, which did not disappoint. 

While Leonardo stuck to a more traditional-looking biryani, it generated a healthy portion that just invites you to grab a seat at the table and tuck in. It garnered extra points in my book for the nice composition of the image as the model included a couple of side dishes including a dhal.

3. Yakitori by Ideogram

Next up was the Toronto-based text-to-image generator Ideogram which was asked to recreate Yakitori, a Japanese dish usually made by grilling skewered chicken. 

Its approach can be described as one that favors simplicity or very strict portion control – nevertheless, aesthetically Ideogram did not disappoint. While it seems to be the only image generator that made an effort to add something extra to the dish, it also needlessly added an extra skewer. Something I would have happily forgiven had it come along with more pieces of chicken.

4. Pizza ai Funghi by Adobe Firefly

I asked Adobe Firefly for its take on an Italian funghi pizza but we also added in the potential extra challenge of asking for a vegan version of the doughy dish. In reality, it could go ahead with vegan cheese toppings but I was curious to see how it would handle this extra bit of information. 

The result was unfortunately the most cartoonish-looking image of the lot even though I specifically requested an AI image that looked like a regular photo. The pizza itself also doesn’t seem too inviting.

5. Tacos Al Pastor by DALL-E 3

Finally, could DALL-E, the AI image generator powering image creation in both ChatGPT and Microsoft Copilot whip up a delicious plate of Tacos Al Pastor? Sí!

What the image lacked in realism it made up for with a healthy portion of taco fillings. As for the modern twist it opted for eggs served sunny side up and even garnished them with some fresh pepper. The plate is surrounded by everyone’s favorite taco toppings.

Final thoughts

While Midjourney, Leonardo, and Ideogram delivered scrumptious results, an overarching lesson learnt was that simply asking the AI image generators to add a modern twist did not produce the intended effect. 

Nevertheless, if you’re inspired to create your own dish you can describe it in a prompt and have a rough idea of what it could look like presented on a plate.

We'd love to see your own AI food creations. If you want to share them you can email ryan.morrison@futurenet.com.

More from Tom's Guide

• Best food delivery services: Grubhub vs Uber Eats vs Doordash
• The best cast iron skillets 2024: Tested and rated
• How to make images with AI using Leonardo

Since February, Elon Musk has been slowly teasing a collaboration between Grok — the intelligent chatbot created by Musk’s xAI and available on the social platform X — and Midjourney. 

If successful, this could see Midjourney’s powerful image generation capabilities integrated directly into Grok. Now, X users have found more evidence in the source code of Grok that further cements the idea of this partnership.

Information is scarce and no formal confirmation has been posted by either Midjourney or X but developer news service TestingCatalog News found a reference to ‘Midjourney’ in the source code for Grok posted to Github.

Grok + Midjourney — what’s this about?

While the critics of AI-powered social content are many, the fact remains that having a chatbot capable of generating multiple social media posts on any topic using a simple prompt is a lucrative concept for influencers and businesses alike. 

Grok already does that, but with the addition of Midjourney, it could also produce high-quality visuals to support those social media posts.

Using a different AI model to create visuals isn't that different to what OpenAI, Google and Meta do, each using a specific AI image generation model but prompted by their respective chatbots.

In February 2024, Musk was participating in an X Spaces conversation when a user called MishaFitton asked him if Midjourney was going to make an appearance in X anytime soon. “We are in some interesting discussions with Midjourney,” replied Musk. “One way or another, we will enable AI generation on this platform [X].”

It’s hardly a surprise. Elon Musk has often been seen sharing similar views with David Holz, the founder of Midjourney. The two have even supported each other’s claims and predictions publicly on social media and during conferences. Given that Grok doesn’t feature any AI image generation capabilities of its own, an API integration with Midjourney could benefit both platforms.

What are the hints this is coming soon?

On June 20, 2024, an AI-focused media outlet called TestingCatalog News found something in the source code of Grok’s standalone version in GitHub. Taking to their social account on X, TestingCatalog wrote, “It seems like a partnership between X and Midjourney has been achieved.” This was followed by a snippet from Grok’s source code that clearly showed the word “midjourney”.

After that, TestingCatalog wrote in another post: “To add a bit more context: This is what currently exists in the code of the standalone Grok version.” This happened just in time for the full release of Grok 1.5.

While the partnership between Grok and Midjourney seems likely given recent events, nothing has been formally announced by either company. While it’s not unnatural for such things to be kept under a tight lid until the last moment, any predictions made right now might edge closer to speculation.

More from Tom's Guide

• ChatGPT Plus vs Copilot Pro — which premium chatbot is better?
• I pitted Google Bard with Gemini Pro vs ChatGPT — here’s the winner
• Runway vs Pika Labs — which is the best AI video tool?

Want to know how to make a comic strip using Midjourney? Look no further. As someone who loves crafting comics but can't draw to save my life, image generators like Midjourney have completely changed the game. These tools, combined with ChatGPT, can revolutionize the way we create, providing stunning visuals that bring our stories to life without requiring a background in art. 

When using Midjourney, it's important to view yourself not as the artist, but as the director, guiding the prompts to match your vision. While the image may not be perfect at first, practice brings you closer to your ideal vision. 

This combination of tools not only transforms your script ideas into vivid, engaging comic strips but also makes the process incredibly fun to watch unfold. Whether you're a seasoned writer or just starting out, this powerful duo can elevate your comic creation process to new heights. Follow these steps to make it happen. 

The finished results...

I'm genuinely pleased with how the comic strip turned out using ChatGPT and Midjourney. Sure, dialing in all the details took some time, but the payoff is fantastic. The final product is a vibrant, engaging comic strip that effortlessly brings my vision to life. It’s amazing what you can achieve with a bit of patience and these powerful tools.

Mastering AI-prompts can significantly enhance your projects. Midjourney's personalize feature allows for tailored artistic creations, and incorporating fun AI prompts, like integrating a song into your story, can add unique elements to your work. For a deeper dive, compare the capabilities of Adobe Firefly vs. Midjourney to take your creations to the next level.

Midjourney's AI-generated images can be incredibly impressive and practical. From copying a character's face to even creating anime, Midjourney is one of the most powerful AI tools around for design projects. However, when you're looking to cut expenses, a paid subscription might not seem worth it. 

For many users, balancing a budget can make it challenging to justify the cost of ongoing subscriptions, no matter how useful or innovative the service may be. If you're in this situation and need to cancel your Midjourney subscription, we’ve got you covered. 

Our quick guide will help you through the process smoothly, ensuring you can manage your expenses without hassle. 

Quick steps

• Log into your Discord account
• Go to Manage Subscription 
• Click Cancel Plan
• Select Confirm Cancellation

If you're curious about trying other AI image generators like Ideogram and Adobe Firefly, be sure to check out our comprehensive guide where we compare their performance by having them create a picture of the perfect day out. 

For a broader overview, take a look at our list of the best AI image generators of 2024.  

Midjourney just made it easier to generate AI images the way you want them with a new ‘personalize’ feature. Using it overrides the style biases the AI model developed in training.

In most cases, there won’t be much difference, especially if you’re not a particularly heavy user of Midjourney — but those putting in the time to generate images can make something unique.

Before you can use Personalize you need to work through an image ranking process, where you compare two Midjourney-generated pictures and pick your favorite. This trains your personalization algorithm and feeds into the image-generation process.

Once you’ve trained up your personalization model you can then add --personalize to the end of your prompt and it will generate based on the preferences you selected while ranking.

Creating prompts to test Personalization

To put Personalize to the test I came up with five varied prompts and ran each of them twice, once with and once without the setting. What it showed is I think I need to spend more time ranking images to further fine-tune my algorithm.

If you are using Midjourney web it is just a switch available under the create settings at the time of writing your prompt. If you use Discord you need to add --personalize to the end. This will also generate a code you can share with others so they can try out your own personal style.

1. Vibrant Vitality

First up I wanted to see if I could get it to create a smartphone-style photograph of something most people wouldn't be able to resist photographing — a beautiful market in a small town. 

The prompt: “An ultra-realistic candid smartphone photo of a bustling, open-air farmers market in a quaint, European town square, with vendors selling colorful, fresh produce, artisanal cheeses, and handcrafted goods, as locals and tourists alike browse the stalls and engage in lively conversation, captured in natural, diffused sunlight with a shallow depth of field.”

I think the personalized image captured the idea of it being a smartphone photo better but I think the default image looks nicer.

2. The Living Room Daydream

I went surreal with the next image, creating something inspired by René Magritte, the Belgian surrealist. Here I wanted to show a cloud in a vintage room.

The prompt: “A surreal, Magritte-inspired scene featuring a giant, fluffy, white cloud floating inside a dimly lit, vintage 1950s living room, casting a soft shadow on the floral-patterned wallpaper and mid-century modern furniture, while a small, red toy airplane emerges from the cloud's surface, leaving a trail of miniature, puffy clouds in its wake.”

Here the default was much closer to something you'd see in Magritte's work but I actually prefer the way the personalized version looks.

3. Wisdom of the Ages

This one was meant to be more of a traditional journalistic photo you might find in a magazine like National Geographic. I think both were amazing but I preferred the default as it was friendlier.

The prompt: “A stunningly realistic, golden hour portrait of a wise, elderly Native American man in traditional attire, with weathered, wrinkled skin, kind eyes, and a gentle smile, sitting cross-legged on a hand-woven blanket in front of a majestic, Southwest desert landscape, captured with a soft, warm light and a creamy, bokeh background.”

4. Aquatic Labyrinth of Wonder

What's weirder than an Escher painting, particularly one based around the concept of a labrynth. Lets put it underwater and see which is furthest from the real thing.

The prompt: “An avant-garde, surrealist underwater scene featuring a school of brightly colored, geometric fish swimming through a submerged, M.C. Escher-inspired labyrinth of impossible staircases and portals, while a giant, translucent jellyfish hovers above, its bioluminescent tendrils casting an eerie, blue glow throughout the dreamlike, aquatic realm.”

Both came up with a great concept but the one in my own personalization feed was better for my tastes.

5. Island Oasis

Next, a simple aerial photograph of a stunning tropical paradise. Both did amazingly and I couldn't decide between them. There was minimal differences.

The prompt: “A photorealistic, birds-eye view of a remote, tropical island paradise, with pristine, white sand beaches, lush, verdant rainforests, and a vibrant, turquoise lagoon, captured from a high-altitude drone during a breathtaking sunset, showcasing the island's rugged coastline, coral reefs, and the vast, shimmering expanse of the surrounding ocean.”

More from Tom's Guide

• ChatGPT Plus vs Copilot Pro — which premium chatbot is better?
• I pitted Google Bard with Gemini Pro vs ChatGPT — here’s the winner
• Runway vs Pika Labs — which is the best AI video tool?

AI Image generators have become essential for anyone looking to boost their visual content. These smart tools can turn ordinary photos into striking works of art with minimal effort. Midjourney, often considered a pioneer in generative AI models for visual content, is a standout in the market. 

The capabilities of best AI generators like Midjourney have skyrocketed in the last few years. Initially limited to generating low-resolution human figures, they can now produce high definition images so realistic that they are almost indistinguishable from actual photographs. 

To begin, ensure that you've signed up on Midjourney's website and log into Discord. We'll guide you through the next steps

Uploading an image in Midjourney

Midjourney is brilliant for opening up new creative possibilities. To maximise your results, check out these valuable tips for getting the most out of Midjourney. 

For those aiming to create cohesive and recognizable characters, the consistent character feature is a must see. And if you're eager to experiement with various aesthetics, these amazing styles will help bring your AI images to life. 

Midjourney and Ideogram are two of the leading and most photorealistic AI image generation platforms on the market. They have similar features and similar styles and both feature in my list of the best AI image generators.

Ideogram is much younger than Midjourney but has quickly cemented its place. Both claim the ability to produce legible text and can create a variety of image styles.

To put those claims to the test I've put them head-to-head in a series of seven prompts, including some requiring legible text, to create my next phone wallpaper.

Creating the image prompts

I set out to create 7 relatively simple and generic prompts that I could use with both Midjourney and Ideogram to see what they came up with. In each case both produce four images and I picked the one I felt was nicest of the four.

In each case I set the aspect ratio to 9:16 which is fine for Android devices but a little thin for the iPhone — but it should still work fine.

1. Abstract Geometric Patterns

The prompt: “A vibrant and colourful geometric pattern with various shapes and sizes. Think of overlapping circles, triangles, and squares in bold hues. The design should have a modern, artistic feel with a sense of depth and movement.”

Here both were good and surprisingly different considering they were from the same prompt. There was something about the Midjourney image that captured my imagination. 

Winner: Midjourney

• Download the Midjourney wallpaper
• Download the Ideogram wallpaper

2. Nature's Beauty

The prompt: “A high-resolution close-up of a dewdrop on a leaf, capturing the intricate details and reflections within the droplet. The background can be a soft blur of lush green foliage, giving it a serene and calming effect.”

Midjourney was an easy winner here. The leaf design was better, it captured simplicity while also maintaining realism. 

Winner: Midjourney

• Download the Midjourney wallpaper
• Download the Ideogram wallpaper

3. Space and Astronomy

The prompt: “An artistic rendering of a cosmic scene, such as a spiral galaxy with swirling stars and nebulae. Include planets and perhaps a shooting star, creating a sense of wonder and infinite possibilities.”

This was close as I think Ideogram was closer to the concept of an artistic rendering. I kept it deliberately vague to see how it handled the confusion and I liked how Ideogram handled it — but the Midjourney image was simply stunning.

Winner: Midjourney

• Download the Midjourney wallpaper
• Download the Ideogram wallpaper

4. Retro Gaming Theme

The prompt: “A nostalgic 8-bit style wallpaper featuring classic game elements. Pixelated characters, coins, and landscapes reminiscent of old-school video games like Super Mario or Pac-Man can add a fun and playful touch to your screen.”

This was a very close one between Ideogram and Midjourney as I liked both images. In the end I came down on Ideogram simply because the design was closer to the concept in the prompt. It created a Pac-Man style layout with Mario imagery.

Winner: Ideogram

• Download the Midjourney wallpaper
• Download the Ideogram wallpaper

5. Minimalist Quote

The prompt: “A clean and elegant design with a motivational quote in stylish typography. The background can be a solid colour or a subtle gradient, making the text the focal point. The quote should be: "Believe in yourself and all that you are." This can inspire or bring a smile every time you unlock your phone.”

The only thing it had to get was a clean design and a correctly rendered text. Midjourney failed on both accounts here and was even over literal with the phone idea. Ideogram not only created a simple elegant design but the quote was spot on.

Winner: Ideogram

• Download the Midjourney wallpaper
• Download the Ideogram wallpaper

6. Whimsical Doodle Art

The prompt: “A playful and whimsical wallpaper filled with hand-drawn doodles and illustrations, such as stars, clouds, and cute animals. The design should be light-hearted and fun, with a quote in the centre in a casual, handwritten font: "Embrace the magic in everyday moments."”

The Midjourney image was closer to the playful and whimsical concept but that isn't to say Ideogram failed, in fact it captured the idea of doodling well. In the end it was the rendering of the quote that won it for me and Ideogram was legible.

Winner: Ideogram

• Download the Midjourney wallpaper
• Download the Ideogram wallpaper

7. Vintage Travel Poster

The prompt: “A retro-inspired travel poster design, showcasing an idyllic destination like a beach, mountain, or famous cityscape. The colours should be warm and inviting, with a vintage feel. The quote in a classic, elegant font can be: "Adventure awaits, go find it."”

Same as the other image tests with text, Ideogram is miles ahead of Midjourney when it comes to rendering text on an image. However, I think the Midjourney image was more stylish and a better overall design. But the text was so bad it lost it.

Winner: Ideogram

• Download the Midjourney wallpaper
• Download the Ideogram wallpaper

Winner: Ideogram

If it wasn't for the issues rendering legible text this would have been a win for Midjourney. In most cases the generated images were more stylized, of a higher quality by default and more photorealistic when required.

However, I found Ideogram was better able to both follow the specifics of the prompt and can generate text unlike any other model I've tried. That is why I handed victory to Ideogram, although it was still very close.

More from Tom's Guide

• ChatGPT Plus vs Copilot Pro — which premium chatbot is better?
• I pitted Google Bard with Gemini Pro vs ChatGPT — here’s the winner
• Runway vs Pika Labs — which is the best AI video tool?

Midjourney is one of the best artificial intelligence image generators on the scene, capable of everything from 3D cartoons to vivid photorealism. Its style reference feature adds an entirely new dimension to these images.

There are thought to be hundreds of thousands of different style options available with the --sref parameter, but even Midjourney isn’t sure exactly how many different styles there are or what each one represents

The AI startup is apparently working on a dictionary of the different styles, which would make it easier to adapt an image. Until then, here are 7 different styles I found while playing around in Midjourney — including one that I used to create a black-and-white band video.

If you've never used Midjourney before, check out our guide to getting started. You can try out any of these prompts in the web or Discord versions.

Using style reference in the journey

There are a few different ways that you can find new styles including stories like this, posts from creators on X or just using the random style feature in Midjourney.

I built myself a random number generator and I’ve been using that to try out lots of different references. These are some of my favorites that I’ve used recently.

1. Medieval Market Square (--sref 1135348669)

I’m using the style reference code 1135348669 for the first image. I’ve decided to call this style Whimsical Watercolor. It looks like a watercolor painting done while drunk. 

The full prompt: “A bustling medieval market square filled with colorful stalls, vendors shouting their wares, and townsfolk in period attire, with a majestic castle in the background --ar 16:9 --sref 1135348669”

2. Astronaut's Journey (--sref 1286756982)

This image uses the style 1286756982, which I’ve decided should be called Vivid Retro Illustration. It is an illustration that uses vivid colors and gives off something of a '60s vibe.

The full prompt: “An astronaut floating in the vastness of space, gazing at a distant galaxy with vibrant, swirling colours, while an alien spacecraft approaches from the distance --ar 16:9 --sref 1286756982.”

3. Autumn Zen Garden (--sref 3413738811)

The next image is a stunning picture of a zen garden. For this I used the style reference 3413738811, which I’ve named Ethereal Anime Realism.

The full prompt: “A serene Japanese garden in autumn, with koi fish swimming in a clear pond, vibrant red and orange maple leaves falling gently, and a traditional tea house in the background --ar 16:9 --sref 3413738811.”

4. Rustic Winter Cabin (--sref 1755994711)

This style has been given the name Winter Folk Art. The code is 1755994711, and it almost has something of a Disney-esque feel, particularly the animals.

The full prompt: “A cosy, rustic cabin in the snowy woods, with smoke gently rising from the chimney, warm light glowing from the windows, and a family of deer grazing nearby. --ar 16:9 --sref 1755994711”

5. Coral Reef Exploration (--sref 3520401691)

For this prompt, which depicts an underwater scene, I picked a style that is as close to "generic digital art" as I could find. The code is 3520401691, and I've named the style Digital Painting.

The full prompt: “An underwater scene of a vibrant coral reef, teeming with colourful fish, sea turtles gliding by, and a mysterious sunken ship partially covered in marine life. --ar 16:9 --sref 3520401691”

6. Volcanic Island Eruption (--sref 1240958064)

I decided to call this style Neo-Pulp Graphic. It works really well with the concept of a volcanic eruption. It comes from the style reference: 1240958064.

The full prompt: “A dramatic volcanic eruption on a tropical island, with lava flowing down the mountainside, plumes of ash and smoke rising into the sky, and a nearby village preparing to evacuate. --ar 16:9 --sref 1240958064”

7. Victorian Rainy Street (--sref 4231799027)

This style has been given the name Dramatic Chiaroscuro Illustration, where chiaroscuro refers to the use of strong contrasts between light and dark. It uses the style reference: 4231799027.

The full prompt: “A Victorian-era street scene on a rainy day, with cobblestone streets glistening, people huddled under umbrellas, a horse-drawn carriage passing by, and gas lamps casting a warm glow through the fog. --ar 16:9 --sref 4231799027”

More from Tom's Guide

• ChatGPT Plus vs Copilot Pro — which premium chatbot is better?
• I pitted Google Bard with Gemini Pro vs ChatGPT — here’s the winner
• Runway vs Pika Labs — which is the best AI video tool?

Midjourney is a powerful platform, but it hides too many of its impressive features behind random numbers or obscure parameters. The learning curve is intense, but as it emerges out of Discord and to its own platform there is one killer feature it needs to create — a style catalog.

Currently, there are many different ways to apply style to an image in Midjourney and one of the best is referencing another image or using the –sref parameter, appending a random number.

Each number corresponds to its own style — as I demonstrated in a recent test — and with tens of thousands of potential styles, finding what works is a constant round of trial and error.

Help might be at hand though. During recent office hours on Discord, Midjourney founder David Holz said they are considering adding a style catalog to the website.

What is Style Reference?

Essentially, Style Reference defines the way the AI image generator interprets the look and feel of the image and is particularly helpful if you can’t think of a detailed or descriptive prompt idea.

Nick St Pierre, a Midjourney expert has created a style guide of his own using AI to automatically tag and label images by color, style or lighting and find related codes. The codes can then be a shortcut to replicate that style of image in the future.

One way to use Style Reference is by sharing the URL of a previously generated image. But having access to a library of what every code looks like will be a game changer for Midjourney.

Moving out of Discord

New features like a reference catalog are all part of Midjourney moving out of Discord. It has been doggedly stuck in the platform since its launch, and it was a good way for the bootstrapped company to get ahead and keep user management costs down.

It recently launched its own web interface, making it accessible to anyone who has created at least 100 images in the Discord bot. This includes the ability to customize image prompts using buttons, toggles and text input rather than parameters with more random numbers.

It isn’t clear yet how a style catalog might work but it could be in the form of a button in the prompt options, opening a new window showing different styles and allowing you to select the best one for your image idea. Right now aesthetics on the web are controlled by three sliders — stylization, weirdness and variety.

You can of course still use parameters in the web version, so if you've found a number you like just add --sref to the end of your prompt like you would in Discord. But going forward it feels like applying styles may be about to get a bit more straightforward.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breaktrhough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

Midjourney has many hidden features, from the ability to copy the face of a character from one image to another through to fine-tuned versions of the model purely for creating anime.

One of my favorite of those features is Style Reference, accessed using --sref. You can do this in three ways — with a number, a URL, or by typing the word random at the end.

Essentially, Style Reference defines the way the AI image generator interprets the look and feel of the image and is particularly helpful if you can’t think of a detailed or descriptive prompt idea. 

The number is tied to the --sref random approach, where a random number is generated and you can then use that number to replicate the style of the image. This is great if you want to create pictures for a marketing campaign or even draw a comic book.

Testing Style Reference on Midjourney

To put this to the test I wrote a script that created a prompt and a dozen random numbers, then I ran those prompts through Midjourney to see just how weird — and different — I could make the images it generated. The answer is surprisingly different.

The script I’ve written takes a simple input — such as “a flying car” — and uses Meta’s Llama 3 running on the Groq Cloud to turn it into a fully formed prompt. I’ve done this to add a greater degree of randomness to the final prompts.

A nice cup of tea

Without wanting to be stereotypically British, I’ve decided to start this test with a nice cup of tea. I fed my script that and it gave me a prompt invoking a wooden table, crackling fireplace, curtains and a cup of Earl Grey tea. I used this prompt on three style references.

The first Style Reference was --sref 649793, which created a pink-hued scene of utter serenity. It's an impressive image with a photorealistic person in a more cartoon-like style.

The second Style Reference was --sref 984728. This one was weird as it seemed to combine a photograph with a hand-drawn sketch. I’ll be using this style again.

The final of the three Style References was --sref 343794. This one strayed furthest from the prompt but was my favorite image. Ethereal drawing of a sleeping woman, tea on the table looking like she has not one care in the world.

A new product launch

Now we’ve got the three styles we’re going to see how they perform on different image prompts, allowing us to compare the style itself rather than the prompt. 

Next up I put the phrase “a new product launch” into my script and it offered up a vision of a group of entrepreneurs and influencers gathered around a pedestal.

We now see the value of the first style (--sref 649793) as applying the neon, high-energy pink hue to a group of people standing around a lit-up podium makes more sense.

This next style (--sref 984728). The premise of the prompt was to show a product launch, which makes me think of Steve Jobs in a black polo neck in front of a large screen. This one was more ghosts in the machine. I like the look but it didn’t work without a strong subject focus.

Finally --sref 343794, our ethereal dream scene. Rather than a product launch this was more pseudo-religious, with people praying to a giant orb in the middle of a room.

Proud to be an American

For the final image, as I started with Tea, I decided to see what my script made of the phrase “proud to be an American." The script is calling Meta’s Llama 3 LLM running on the Groq Cloud, taking the phrase and turning it into a more descriptive idea. 

For this idea, it suggested an image of a warm summer evening in a small town where friends are gathered in a backyard adorned with twinkling stars and a vintage flag.

I think the third style will be the winner for this scene but first up our neon inferno (--sref 649793). It worked better than I expected, although it gives off GTA vibes.

Now let's try the sketch-inspired photograph (--sref 984728). With this prompt, it was less photograph and more illustration used as a cut-scene in a biopic about a country music star.

Finally our ethereal scene (--sref 343794). I was right to predict this would handle it best, although I’d prefer if it was a little less washed-out looking.

Conclusion

You can take any of the three styles we tried in this test and apply it to your own prompts in Midjourney. Even if you only give it a one-word prompt — summer — the inclusion of style reference will give you an interesting image.

If you don’t want to try any of the numbers I’ve used, just add the word random after --sref and it will give you a random number. Once generated you’ll be able to see the number and if you like it use it in future prompts. If you don’t — type random again.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breaktrhough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

Midjourney has been building out a new web app for creating artificial intelligence images for a few months and it is finally becoming more widely available.

Until now you’ve needed a Discord account to wrestle with the Midjourney AI generator bot. While this does work, and you get a lot of control over the final generation through parameters, it isn’t the easiest or most user-friendly experience.

Anyone who has created at least 100 images using the Discord-based Midjourney bot can now access the web-based tool. This includes new “rooms” to discuss different aspects of AI image generation. This feature is similar to the Discord server and the NightCafe AI platform.

I tried it out for myself and while it might be a bit slower than Discord, and require learning a new UI, it is also significantly better than the previous experience. I’m not going back to Discord.

How the new Midjourney web interface works

You will still need a Discord account to sign in to the Midjourney website and get your account for the AI image generator. Once you've signed in you'll be presented with an array of impressive images created by other users, and a big text box to generate an image yourself.

In multiple locations Midjourney stresses that this is an “alpha” feature, meaning that it is not ready for general, mainstream use and could be full of bugs. A beta version is likely soon.

To get started you can simply type what you want to see in the big box at the top, hit enter and wait for it to make your images. But the biggest benefit of the web UI is easier customization.

You no longer need to remember dozens of parameters and how to use them properly, ensuring you type out the – and get the order right. Now you just select from a drop-down menu.

The full suite of potential parameters haven’t been transported to the Midjourney web interface yet — they do stress it is an alpha build — but that will change over time and for now you’ve got more or less everything you need to make a compelling AI image.

Why is Midjourney moving to the web?

Midjourney has remained doggedly committed to Discord since it first launched two years ago. It is a great platform for a company to get started, allowing bots and integrations as well as handling all user management and interactions at scale — cheaply.

Runway and Pika Labs also started in Midjourney, PixVerse and others still use it and every major AI platform from big players like Adobe to smaller startups like FinalFrame run servers. But as a company grows they usually build out their own infrastructure, not Midjourney.

The problem with Discord is that it isn’t great for enterprise users, it can hamper new user acquisition as it not only requires a Midjourney learning curve but also a Discord one and Midjourney will have limited access to data on users and their behaviour.

Moving to the web puts the entire process in the hands of Midjourney. It will also allow them to build out more unique features and make it easier for users to get started without having to register for Discord. My favorite feature though is being able to toggle a switch, tick a box or select from a menu rather than type – and remember the right number.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breaktrhough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

Midjourney has finally updated its random style feature so that it works with the current version of the artificial intelligence image generation model. 

Added as a parameter at the end of your prompt, it effectively leaves the look and feel of the image entirely up to the AI to determine. 

During testing I had the same prompt create everything from anime and action figures to full on film noir and greyscale. Its an easy way to test the possible.

If you haven't used Midjourney before and want to get a better feel for the powerful AI image generator we have a guide to using Midjourney and a selection of prompts to try out.

How does Midjourney random style work?

You can randomize the style of an image by typing --sref random at the end of your prompt. If you repeat the prompt multiple times with the same parameter you’ll get a range of images in a variety of styles. It didn't duplicate once in a dozen tests.

This is a great way to get started with Midjourney if you’re new to the AI image generator as it gives you an idea of what is possible, how styles can vary and inspiration for your next image.

If you like a particular style you can then type --sref and the url of the image you liked after another prompt to keep that style across future images.

Putting Midjourney random style to the test

I came up with a simple prompt that AI image models sometimes struggle with — a cat in a spacesuit on the moon — and then added the –sref random parameter.

David Holz, Midjourney CEO said that this change was part of a test of new algorithms for "exploring the latent space of visual styles." He described --sref random as a "fun tool for everyone play with."

"This is meant to be a test of a underlying algorithm that we make more advanced tools for in the future," he declared.

I ran each generated image through Midjourney Describe to get an idea of how the AI model interpreted the style it referenced. What I'd like as a future upgrade is a way for Midjourney to also offer up a description of the style when you use random.

1. Grumpy Cat (photorealism)

Midjourney describes this image as: "A Persian cat in an astronaut suit on the moon, photo realistic, high resolution photography."

This captivated me the moment in generated but certainly isn't a style I'd bother to replicate as its basically a generic photorealism, albeit with an element of tilt shift.

2. Duke Cat-em (Sci-fi)

Midjourney describes this as: " A cat astronaut in space, holding an energy weapon and wearing full protective gear with glowing eyes against the backdrop of Earth, space background, detailed character design in the style of space punk, high resolution, high quality artwork."

This one feels like something out of a game adaptation of a dark sci-fi comic book called something like Space Cats and the Battle for Planet Earth. I would definitely try this style again.

3. Puzzled Cat (3D render)

Midjourney says: "A Bengal cat in an astronaut suit on the moon, in the digital art style, a full body portrait, a hyper realistic illustration, with detailed background elements, cinematic lighting, vibrant colors, with elements of fantasy and surrealism, high resolution, high detail, an octane render, created in 3D using the Unreal engine."

My first impression of the image was of something out of a classic sci-fi show that had been colorized. The retro-futuristic space suit and slightly vacant stare are my favorite parts of the image. There is something enchanting about this style.

4. Menacing Cat (Ink Wash)

Midjourney described this image as: "A black and white drawing of an astronaut cat on the moon, full body, with a detailed background, hyperrealistic illustrations in the style of detailed character designs, ink wash paintings in the fantasy art style."

This is an image I'd like hanging from my office wall but isn't necessary a style I'd try again. Its a bit intense for my taste. But I'd consider trying the style on a different topic.

5. Surprised Cat (Meme)

Midjourney described this as: "A white cat on the moon looking at earth in space, screencap from a YouTube video about cats that went to space in the style of cats that go to space."

While it does have something of a YouTube thumbnail about it, this image screams meme to me. There's even space for the text. I'd put something like "how much did I have to drink last night" in the top left corner. I'd use the style again to make a meme.

6. Cat-astrophy (Glitch)

Midjourney didn't have much to say about this one, simply describing it as: "A cat in an astronaut suit on the moon, black and white photography, glitch art."

I love the intensity of this image. Not only does the picture feel broken but there is a pain in the face of the cat as if it is pondering the meaning of life. I'd use this style again in a heartbeat. 

7. Cute Cate (Kawaii)

Finally Midjourney writes: "A 3d render of an orange and white calico cat in space suit on the moon, cute cartoon character design with little details, walking towards camera with astronaut hangar in background, cute kawaii."

One word — adorable! This is a fun style that could be used to create images for an eventual stop-motion animation in Runway or Pika Labs.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breaktrhough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

I create a lot of content with generative AI, and at this point it feels less like a tool and more like part of my operating system. On most mornings, I’m bouncing between models instead of apps — generating visuals, refining them, animating scenes and layering in voice, all before my second cup of coffee.

My workflow changes constantly because the tools evolve almost weekly. One day I’m building images, the next I’m turning them into motion clips, and by the afternoon I’m using AI voices or music generators to finish the piece. The process that once required a full creative team can now happen in a single browser tab.

The speed of improvement is staggering. Just a few years ago, AI image generators struggled with hands, text and anything resembling realism. In 2026, the latest models produce cinematic lighting, consistent characters and photorealistic detail that can pass casual inspection — and they do it in seconds.

What used to feel experimental now feels production-ready.

In 2026, the AI image race isn’t just about who has the most powerful model. What separates platforms now are the creative tools wrapped around the generator — editing canvases, in-painting controls, upscaling, character consistency tools and built-in workflows that turn a single image into a finished asset.

Features like canvas editing environments, precision retouching, high-resolution enhancement and advanced editors that can modify external images have transformed these tools from simple prompt engines into full creative studios.

Choosing the right platform isn’t easy — especially when premium tiers promise better quality, faster rendering and commercial-ready outputs. That’s why I didn’t just try these generators. I pushed them hard.

I stress-tested each model with a wide range of prompts and real-world use cases to see how they perform beyond the demo images. I evaluated photorealism, text accuracy, character consistency and overall usability. I also tested how well the images held up in motion workflows, including animation pipelines, and compared how convincingly each generator renders human subjects.

The results form the foundation of this guide to the best AI image generators you can use right now.

How we test AI image generators

The best way to evaluate any AI tool is simple: use it constantly and push it beyond the marketing demos. I generate images every day as part of my workflow, rotating through the major platforms in this guide and running side-by-side comparisons to see where each one shines — and where it breaks.

I test with varied prompts, complex compositions and real-world scenarios, then push outputs through editing, animation and video pipelines to see how well they hold up across modern creative workflows. Today’s image generators don’t live in isolation; they feed into motion tools, lip-sync engines and multimodal production pipelines. Understanding how they perform end-to-end matters as much as the initial render.

Each platform also responds differently to prompting. Some reward concise direction, others thrive on cinematic detail or reference images. Part of my testing focused on uncovering the prompting approaches that consistently deliver the best results.

For this guide, I focused on platforms rather than standalone models. Many leading models — including Gemini, Grok, Stable Diffusion variants and others — now appear across multiple services and creative hubs. At the same time, proprietary models remain exclusive to certain ecosystems. The lines are blurring as APIs and integrations make top-tier models accessible across creative platforms.

To keep the comparison consistent, I developed a set of standardized prompts designed to test realism, text rendering, composition and character fidelity. The resulting images appear in galleries for each section so you can compare outputs side by side and see how each platform performs under identical conditions.

Best AI image generator overall

ChatGPT stands out as the overall best image generator. Different than Sora, the image generator is built in to the model, allowing users to text an image prompt within the chat.

OpenAI's model interprets detailed prompts with precision, producing images that closely match user descriptions. In comparative tests, it outperformed competitors like Midjourney V7 in areas such as photorealism and complex scene recreation.

Unlike many AI image generators, including previous versions of ChatGPT, ChatGPT-5 effectively incorporates legible and accurate text within images.

Leveraging its knowledge base, ChatGPT-5 can generate contextually rich images without needing exhaustive prompt details. The simplest text prompts can yield accurate depictions without further explanations.

Users can also iteratively refine images through natural language instructions. Whether it's adjusting colors, adding elements, or modifying styles, ChatGPT-4o facilitates seamless edits within the chat interface. The ease of which edits can be made also makes ChatGPT a top bot in the catagory.

Finally, the newest ChatGPT-5 features is an Image Library, allowing users to easily access, manage, and revisit their generated images.

ChatGPT-5 combines advanced image generation capabilities with user-friendly features and ethical considerations and strict safety measures in place.

Honorable mention for best overall image generator

Nano Banana is the collective name for Google Gemini’s native image generation and editing capabilities. It is a state-of-the-art suite capable of text-to-image creation, image-to-image editing, and multi-image composition.

Directly through Gemini, either on desktop or in the app, users can use conversational, natural language to iteratively refine and adjust visuals. Every generated or edited image automatically includes an invisible SynthID digital watermark to clearly identify it as AI-generated and promote transparency.

Nano Banana (The "Fast" Model): It is optimized for high-volume, low-latency tasks and quick, casual creativity. The AI image generator excels at maintaining character consistency, allowing you to keep the same person or product recognizable across multiple different scenes. It supports intuitive local editing, meaning you can use simple text prompts to seamlessly combine photos, change a subject's pose, or remove specific background objects.

Nano Banana Pro (The "Thinking" Model): This professional-grade tier is built on the advanced Gemini 3 Pro Image model. It acts as a reasoning engine that utilizes a "Thinking Process" to plan out complex physics, object relationships, and lighting before rendering the scene to deliver native 4K resolution output for incredibly crisp, high-fidelity professional assets.

It features superior text rendering capabilities, allowing you to generate perfectly spelled, legible typography in multiple languages directly onto your images. The tool offers advanced real-world knowledge and spatial understanding, making it highly effective for creating accurate infographics, UI mockups, and data visualizations.

To be completely candid, while both models are designed to handle ongoing conversational edits, AI image generators can sometimes get "stuck" if you ask for too many complex changes in a single chat. If you find the model ignoring your requested edits, simply starting a fresh conversation with your current image and a new prompt is a highly reliable workaround.

Midjourney remains one of the most recognizable names in AI image generation, even as the market has expanded into full creative platforms with editing suites, animation pipelines and multimodal workflows. While some competitors bundle multiple models and built-in video tools, Midjourney continues to stand out for its visual quality, artistic control and distinctive aesthetic.

The platform has evolved well beyond its early Discord-only roots, though its interface and parameter-driven workflow still require a learning curve. Once you understand how to guide it, however, Midjourney offers exceptional control over composition, lighting, style and mood. Power users routinely produce images that rival professional photography and high-end concept art.

One of Midjourney’s enduring strengths is how precisely users can steer outputs. Style references, character consistency tools and parameter controls allow creators to guide visual direction with remarkable accuracy. This flexibility makes it especially valuable for branding, storytelling and iterative design work.

Recent updates have pushed Midjourney beyond still imagery. The rollout of Version 7 introduced faster draft rendering, improved realism and personalization features that adapt outputs to a user’s visual preferences over time. The company has also expanded into motion, with its V1 video model capable of turning prompts or still images into short animated clips.

Looking ahead, Midjourney continues to explore deeper creative capabilities, including enhanced motion workflows and dimensional outputs, as the industry moves toward fully immersive media creation.

At the same time, the platform — like much of the generative AI space — is navigating ongoing legal scrutiny around training data and character likenesses. How these cases evolve could shape the future of AI image generation across the industry.

Despite growing competition, Midjourney remains a go-to tool for creators who prioritize visual quality and fine-grained artistic control.

Best for Features

Leonardo began as a polished interface for Stable Diffusion models, but it has quickly grown into one of the most versatile AI art platforms available. The launch of its Phoenix model marked a major step forward, delivering photorealistic results that rival some of the best outputs in the space, while still offering extensive style controls and fine-tuned model variants.

In mid-2024, Canva acquired Leonardo and began integrating its technology into Canva’s Dream Lab, an AI image generator designed to produce everything from cinematic landscapes to hyper-real portraits. Dream Lab runs on Phoenix while also drawing from Leonardo’s expanding model lineup, including PhotoReal for lifelike imagery, Lightning XL for faster renders, and Lucid Origin for bold, stylized visuals. Users can also experiment with Elements — fine-tuned styles like sketch or sculpture — and apply themes such as cinematic, food or long exposure before generating an image.

Since Canva’s acquisition, Leonardo has evolved far beyond a standalone image generator into a core engine powering a broader creative workflow. Its Phoenix model continues to improve realism, prompt accuracy and text rendering, while new style reference and brand consistency tools make it easier for teams to produce cohesive visual content at scale.

At the same time, Leonardo’s real-time canvas, sketch-to-image controls and advanced editing features support iterative design, and expanding motion and video capabilities reflect the industry’s shift toward multimodal creation.

Together, these updates position Leonardo not just as an AI art tool but as part of an end-to-end content production ecosystem built for marketers, designers and creators.

Best for text on images

Ideogram has long been a go-to for anyone who needs AI-generated images with flawless text, and in 2025 it’s only getting better. Originally known for design-friendly tools like color palettes, a dedicated “design” style, and text rendering that puts rivals to shame, the platform now offers even more creative control.

Its Canvas editor remains a standout, letting you refine or completely rework images with extended text prompts, perfect for fixing text alignment or adjusting graphic design elements. Batch Generation streamlines workflows by creating multiple images at once, making it easy to spin up posters, product mockups, or social media graphics in bulk.

Recent updates have taken things further. The new 2a model improves speed and cost-efficiency for design and photography workflows, while Ideogram 3.0 adds sharper photorealism and a style reference system that lets you upload up to three images to guide the look and feel of your results. You can also browse and apply from over 4.3 billion style presets using style codes.

Mobile users aren’t left out either, Ideogram’s official iOS app brings all these capabilities to your phone, with an Android version on the way. Combine that with Magic Prompt, which automatically refines your text for better results, and you’ve got a tool that’s equally suited to professional designers and casual creators.

Whether you’re building a marketing campaign, creating a movie poster, or mocking up a greeting card, Ideogram makes it easy to get polished, text-perfect images with minimal fuss, and that’s why it’s still one of the best AI image generators you can use today.

Best for collaboration

Google’s Gemini chatbot has had a rocky history with image generation. Early versions, powered by DeepMind’s Imagen 2, ran into trouble in 2024 when the system produced inaccurate depictions of people, prompting Google to suspend the feature entirely. Now, after a major overhaul, it’s back — and it’s much more capable.

The current system runs on Imagen 4, Google’s latest AI image model, which delivers sharper visuals, more accurate text rendering, and improved realism across styles — from photorealistic portraits to oil-painting textures. It also fixes the earlier issues with generating people, with stronger safeguards in place to avoid the same missteps.

Gemini’s image tools now go beyond simple generation. You can edit existing AI-generated or uploaded images by changing backgrounds, swapping elements, or refining specific details with natural language prompts. For developers, Imagen 3 is also available through the Gemini API, with fine-tuning options like aspect ratio control, multi-image outputs, and built-in SynthID watermarks for authenticity.

These upgrades are available to Gemini Advanced, Business, and Enterprise users, making them accessible to a much wider audience. Combined with its faster generation times and integrated editing tools, Gemini is evolving into a serious contender in the AI art space — and a far more polished one than the version that stumbled out of the gate.

Best for interaction

Image generation inside ChatGPT has evolved quickly, and it’s no longer tied to a single model or paywall. Free users can now create a limited number of images each day, while paid tiers offer higher limits and faster generation. You can generate visuals directly within any chat, making it easy to move from an idea to a finished image without switching tools.

ChatGPT now uses OpenAI’s latest multimodal image capabilities rather than relying on a standalone model experience. This means you can describe an image conversationally, refine it with follow-up prompts and iterate in real time — all within the same thread.

The experience is designed to feel collaborative. You can ask for adjustments, change styles, improve realism or tweak details like lighting, composition or text. The system is especially strong at rendering readable text, maintaining stylistic consistency and producing photorealistic or stylized images on demand.

Editing remains one of the biggest strengths. Select an image, highlight an area and describe what you want changed. Whether you’re swapping backgrounds, correcting text, or adding new elements, iterative edits feel natural and fast.

While image generation models continue to improve behind the scenes, the real advantage inside ChatGPT is the conversational workflow: you describe, refine and evolve visuals without starting over.

Best for ethical training

Adobe Firefly has always stood out for its ethical approach to AI, its training set is almost entirely built on Adobe Stock images, giving creatives more confidence that their work is copyright-safe. Adobe even offers financial indemnity for images generated with Firefly, something few rivals can match.

Now, Firefly is expanding well beyond still images. The new Firefly Video Model, currently in public beta, lets you generate high-quality, IP-safe video clips from text prompts or reference images and send them straight into Premiere Pro. It works hand-in-hand with Generative Extend, a feature that can seamlessly lengthen shots while keeping the style and motion consistent.

On the image side, the Image Model 4 upgrade delivers sharper detail, more accurate prompt interpretation, and even better realism for both people and environments. Adobe has also rolled out a Firefly mobile app for iOS and Android, putting image and video generation, plus Photoshop-style Generative Fill and Expand, in your pocket. Creations sync automatically to Creative Cloud, so you can start on mobile and finish on desktop.

For collaboration, Firefly Boards is now in public beta, offering an AI-powered moodboarding and remixing space that supports not only Adobe’s own models but also third-party tools like Luma, Ideogram, Runway, and Pika.

With these upgrades, Firefly has evolved into a full creative engine, spanning ideation to production, while holding on to its biggest advantage: giving professionals cutting-edge AI tools without sacrificing commercial safety.

Best for community

NightCafe was one of my first introductions to the world of generative AI. It was initially built on some custom models and later expanding to use Stable Diffusion-based models that have since been fine-tuned and customized.

As well as offering access to a wide variety of models, including Flux, Google's Imagen 3, Stable Diffusion, OpenAI's DALL-E 3 and now Ideogram, NightCafe makes it relatively easy to fine-tune or train your own models and lets easily manage and share your creations.

It even allows you to animate the individual photos using Stable Video Diffusion and upscale images using a range of upscaler models. You can even animate with Runway Gen-3 Turbo and Luma Labs Dream Machine within the single platform.

Its community is the real secret sauce behind why I keep going back to NightCafe again and again, regardless of which new models or services launch. There are regular competitions, chatrooms and creator jams for new ideas.

The company works to support that community, even engaging in partnerships with third party companies to offer prizes and free credits.

The credit system is also particularly helpful. Other services require a fixed monthly payment or have costly credit packs but NightCafe allows you to buy what you need.

It isn't perfect. The UI can be busy and the initial user journey can be daunting, but once you've engaged with it a while you'll find yourself coming back again and again.

Best for Innovation

Freepik is an interesting platform. Originally know for its library of images, it has very quickly become a major force in the artificial intelligence space. Some of this was driven by its acquisition of AI-upscaler Magnific earlier this year.

The AI generation part of the platform is incredibly easy to use and intuitive. It also includes a prompt enhancer that uses a large language model to improve your simplistic prompts.

You can add filters and elements to the image generation such as custom styles, colors and lighting. You can also specify the camera type if you are taking a photo. When exporting you have an easy upscale button that happens automatically before offering the download.

The feature set goes beyond simple image generation. One of these is sketch enhancement using AI. You can doodle on the screen and the AI will turn your doodle into an artwork or photograph.

Outside of generation, editing and sketching you can also upscale images from outside of Freepik, expand canvas in a similar way to Photoshop and use AI to retouch existing images.

Freepik has become a good all-round AI powerhouse, a way to play with image creation, including offering the latest models. It has all versions of the Flux family of models including Flux 1.1 Pro, the new rapid and realistic version.

The big stand out though is Mystic, built by the Magnific team. This is a fine-tuned and customized version of Flux that is perfect for photorealistic depictions of people.

Best for originality

Luma’s Dream Machine is more traditionally an AI video platform, at least that is how it started. With its most recent update, and the addition of the Photon image model it has become one of the most innovative approaches to AI creation.

The Photon model can generate compelling images and is surprisingly creative and personalizable through complex prompting. It is also higher resolution for a lower price than other similar-sized models from Luma’s competitors. Its big differentiator is its prompt adherence and integration with the new look Dream Machine, which is now both image and video.

In Dream Machine a user enters a prompt and then collaborates with the platform to enhance both the prompt and adapt the image. I tested this in a review and found that the customization makes it very easy to get the exact image you are looking for.

One of my favorite features of Dream Machine is the interface. It is more like a chatbot than a traditional AI image generator. After entering your prompt you get the full prompt as well as four images. Keywords within the prompt are highlighted and can be quickly customized to change specific elements within the image. The Dream Machine platform has undergone a comprehensive redesign to improve user experience.

You can then also opt to show more images, brainstorm on the side, and even reply to the response to that prompt to get specific changes. Clicking reply will load the generated images into the chat box as an attachment where you can send it to the video editor or change the image using as simple natural language prompt.

Luma Dream Machine works by creating “boards” for each topic. You can then iterate through the idea, brainstorming with the AI and keeping it all collected in one location.

Recognizing the diverse needs of its user base, Luma AI has introduced subscription tiers to cater to different levels of usage:

Hobbyist: $9.99 per month

Explorer: $29.99 per month

Professional: $99.99 per month

Enterprise: Custom pricing for larger teams

These tiers offer varying levels of access and support, allowing users to select plans that best fit their creative requirements.

Best for video linkup

Kling’s KOLORS is its dedicated AI image model. A lot of the big video platforms are launching custom image models as they use them to enhance their video platforms. Luma has Photon, Runway is launching Frames and Haiper already has an image version of its model.

KOLORS is impressive for visual consistency, human realism and portrait aesthetics but I’ve found it struggles with text rendering, something other models now largely do very well.

One area it is particularly good is in generating portraits. It captures the light and shadow required from a prompt particularly well and follows a prompt accurately.

KOLORS also comes with a virtual “try-on” where you can generate a virtual model and then either upload or generate an outfit for the model to wear. It comes with a series of example images you can use for inspiration and each image costs 1 credit.

The true value of KOLORS though comes from the deep integration with the Kling AI video model. The user generates an image and then clicks to animate it. Kling is one of the best AI video models currently available. It can create hyperrealistic video with accurate motion.

Final thoughts

Generating AI images is a fun process. You get to see an idea you've had in your head almost immediately and with new features like Canvas you can edit and make changes to that image to fit your exact concept.

I've used Ideogram to create posters to remind the children to turn off lights, artwork for app icons and even entire app designs. I've used Midjourney to create photos worthy of printing and Adobe Firefly regularly contributes to cover images for my stories.

I hope this guide has given you some insight into the best tools for creating AI art. If you've tried something I've not included send me an email, find me on X or leave a comment and I'll give it a review.

More from Tom's Guide

• 7 ChatGPT productivity hacks that you're not using — but should be
• I use Gemini every day — here are 7 mind-blowing prompts I can’t live without
• These 9 AI prompts work like magic — no matter which chatbot you use

Asking someone to describe their perfect day out is a great icebreaker when you want to get to known them better. The response can tell you a lot about their personality, hopes and ideals, without becoming overly personal or intrusive.

With artificial intelligence coming on in leaps and bounds around reasoning, creativity and its knowledge of the world — I decided to put that same question to the leading AI image generators to see what they came up with and how they understand the idea an ideal situation.

I needed to come up with a prompt that was complex enough to ensure consistency but simple enough not to prescribe the ideal day — pushing the AI to interpret what it means.

Surprisingly they all depicted scenes of the natural world, either with friends or depicting scenes of solitude and calm. All but two also feature water, either as a lake or an ocean. 

How does AI see the perfect day

I’ve tried the same prompt — with no additional settings — on Midjourney, Adobe Firefly, Leonardo, DALL-E 3, Ideogram, Meta Imagine and Google ImageFX.

The prompt: “Create an image that captures the essence of a perfect day out. The image should evoke feelings of happiness, relaxation, and enjoyment. Consider the setting, activities, and atmosphere that would make this day truly memorable.”

Where possible I’ve asked it to make the image landscape and in every case I’ve selected the first image of a set, even if it generates multiple images.

Ideogram

While Ideogram might be best known for its text on image capabilities, it is also a powerful AI image model capable of producing photorealistic depictions of a false reality. 

Ideogram understood a “perfect day out” to be one spent with friends on a sunny day getting back to nature. All four images it generated were of people smiling as a group.

The image is evocative and makes you smile. I’d enjoy that day out. I like a walk in the woods with friends on a sunny day — especially if there is a nice BBQ at the end.

Midjourney

Midjourney is also able to generate stunning photorealistic depictions of the world, such as this hammock dangling over the edge of a lake as the sun begins to set in the background.

It is interesting that Midjourney depicted the perfect day out as one of isolation and peace. The other three images it generated in the set of four were all of a similar concept — alone time.

While I’m a fan of the peace and isolation depicted in the image, I wouldn’t want to sit in that hammock as I’d likely fall off and spend the rest of the evening drying off.

Leonardo

Leonardo seems to have taken inspiration from a commercial for a couples' retreat. The image shows a couple sharing a sun lounger sitting by a beautiful blue ocean on a sunny day.

All four images depicted the same environment although one was a woman sitting alone, another a man alone on the chair with a drink and the fourth was just the sun lounger on its own by the sea.

While I don’t enjoy the beach, there is no beach in this image. It seems to be an ocean pool coming right up to a paved platform. I would enjoy this day out, especially with a nice cold drink.

Adobe Firefly

Not every depiction of a perfect day has to be photorealistic. One thing Adobe Firefly is particularly good at is creating artistic scenes. This shows a crystal-clear river in a lush valley with snow-capped mountains in the distance.

The other images were along the same lines and the fourth image depicted a man reading a paper in a chair, sat next to a pet dog while watching people play in the lake.

This to me seems like an amazing day out. A peaceful environment, watching the fish swim in the clear lake and sipping on an ice cold drink reading a book.

DALL-E 3

DALL-E also went down the artistic route with its interpretation of the perfect day out. It captured children playing around a tree, again with a lake and sunshine.

Only two images were created by DALL-E for this prompt and both were interpretations on the same concept — people enjoying time out in nature as a group.

I’ve had many days like this with my children over the years, sometimes with and sometimes without a kite. The picnic on the grass, throwing a ball and laughing together would surely be a perfect day out for many.

Meta Imagine

Meta Imagine only generates square images but we still get the same sense from it as those in widescreen. Here we have a family on a beach enjoying food, sun and music together.

Meta generated the most variety in its images. The other inlcuded boating on a lake, an elderly couple sitting by a creek and a serene view of a river running by a forest.

I’m not personally a big fan of the beach and while I do play guitar I’d never be comfortable playing it on a public space like that. This is anxiety-inducing for me but they look to be having a lot of fun together. 

Google ImageFX

The image generated by Google ImageFX has no people in it and no indication people were every anywhere near it. For Google, the perfect day out seems to be one of nature and emptiness. 

ImageFX generated three images and the other two were just versions of the first with a few style tweaks. I don’t think it understood or followed the prompt assignment that well.

While the environment looks like an interesting one to visit, it doesn’t really show a day out as such, more a place to go for a day out — but it does look stunning.

Which AI image generator won?

Ideogram was my winner in this test. It not only understood the idea of a perfect day out but generated a sense of bliss on the face of the people in the picture.

It went beyond a literal interpretation taken by some of the other AI tools and instead focused on the happiness of the people enjoying the day.

That said, Adobe Firefly captured the closest to my perfect day out. A calm day by the lake with loved ones and a good book.

More from Tom's Guide

• OpenAI's GPT Store is now live with over 3 million custom chatbots to try
• OpenAI launches custom chatbots — this is how they work and what they can do
• Forget ChatGPT — OpenAI now lets you build your own chatbot for any task

Adobe has always sold its artificial intelligence image generator Firefly as a form of ethical AI, training it primarily on its own library of licenced stock images — millions of photos, designs and illustrations from the Adobe Stock library.

A new report by Bloomberg claims this squeaky clean image may not be as perfect as the Photoshop-maker would like us to think. The "commercially safe" model may have a few not-so-well sourced skeletons in its training data.

During the process of training the Firefly AI model, some of the images came from competitor Midjourney. The startup has never declared the source of its training data but many suspect it is from images it scraped from the internet without licensing.

Adobe says only about 5% of the millions of images used to train Firefly fell into this category and all of them were part of the Adobe Stock library, which meant they'd been through a "rigorous moderation process."

Why is this such a big deal?

When Adobe first launched Firefly it offered an indemnity against copyright theft claims for its enterprise customers as a way to convince them it was safe.

Adobe also sold Firefly as the safe alternative to the likes of Midjourney and DALL-E as all the data had been licensed and cleared for use in training the model. 

Not all artists were that keen at the time and felt they were coerced into agreeing to let their work be used by the creative tech giant — but the sense was any image made with Firefly was safe to use without risk of being sued for copyright theft.

Is Firefly still copyright safe?

Despite the revelation some of the images came from potentially less reputable sources, Adobe says all of the non-human pictures are still safe.

A spokesperson told Bloomberg: "Every image submitted to Adobe Stock, including a very small subset of images generated with AI, goes through a rigorous moderation process to ensure it does not include IP, trademarks, recognizable characters or logos, or reference artists’ names."

The company seems to be taking a slightly more rigorous step with its plans to build an AI video generator. Rumors suggest it is paying artists per minute for video clips.

I have reached out to Adobe for a comment on this story.

More from Tom's Guide

• Best photo editing software in 2024
• Adobe AI image generator just went global with huge update — what you need to know
• Adobe just put generative AI into Photoshop — what you need to know

Midjourney is one of the most powerful AI image generators available, capable of producing stunning and highly detailed visuals. However, unlocking its full potential requires more than just typing in a simple prompt.

Unlike other AI tools that rely primarily on natural language or menu-based settings, Midjourney offers a unique system of parameters—short commands that fine-tune the AI’s output. These parameters allow you to control aspects like aspect ratio, style, chaos levels, and even the version of the model being used.

With Midjourney, you can create your own comic strips, anime, and even stickers, opening up endless creative possibilities. Whether you’re aiming for ultra-realistic portraits, abstract art, or cinematic landscapes, knowing how to use Midjourney’s settings effectively can dramatically improve your results.

In this guide, we’ll cover five essential tips to help you get the most out of Midjourney and create images that truly stand out.

Making the most of Midjourney parameters

You can use the parameter options to control everything from the aspect ratio to the version of the Midjourney model you use to create your artwork. I’ve pulled together a few of the more useful with a brief guide to using them.

One thing to note if you are on a Mac — Apple’s operating system will often convert a double dash (--) into an em-dash (—) but it isn’t an issue as Midjourney supports both.

1. Setting the aspect ratio

There are two ways to change the aspect ratio using parameters --aspect or --ar and you can more or less use any ratio as long as they are whole numbers. It defaults to a square image, or 1:1 and if you want it widescreen you'd use 16:9 or 9:16 for a smartphone-style portrait view.

You can change the aspect ratio of an image using the custom zoom option. You can add a different aspect ratio using the --ar parameter in the pop-up box.

2. Character reference

One of the most difficult, but important things to get right in AI image creation is consistent characters. Particularly if you want to use the images in video production. Midjourney does this using the --cref parameter. 

It works best when you use an image generated by Midjourney as the source material, especially if that image only has one single character. Just copy the upscaled image URL and paste it into a new prompt after the --cref parameter. So --cref .

3. Let chaos reign

While Chaos might sound like something extreme, the reality is it just adjusts how much variety is put into the generated image. You use it with --chaos or --c and it influences the variations across the four images generated. It runs from 0-100.

A high chaos value will give four varied and unexpected results, a lower value will create more reliable and repeatable results. If you don't specify a chaos value it will create four images that are very similar such as four near identical shells on four near identical backgrounds. A high chaos value will create four very different images.

4. Being negative

Negative prompting can be hit and miss but Midjourney allows you to specify terms you don’t want to appear in the final generated image. This is accessed using the --no parameter.

For example, if you want to create a picture of a garden that is just a lawn then you might use --no plants, or if you want an empty street adding --no cars could help.

5. Setting some style

Stylize allows you to create images not tied to the default styles for Midjourney which tends to favor artistic color, composition and forms — for example it could be used to create a child's drawing without any flare. It is accessed with --stylize or --s .

If you keep the stylize value low it will more closely match the prompt but be less artistic. Having a very high stylize figure will create a more artistic image but less closely linked to the prompt. The default is 100 but runs from 0 to 1000.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breaktrhough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

AI tools like Midjourney, ChatGPT-5 and DALL-E can be a game changer when it comes to creating content but unfortunately, many of them are locked behind subscriptions or only available in limited access.

There’s nothing hackers love more than something in short supply and according to a new report from Bitdefender, they’ve devised a complex way to use these tools — and access to them — to infect unsuspecting users with info-stealing malware.

Just like with other online scams, this one starts on Facebook before potential victims are led to a malicious site controlled by the hackers behind this campaign. From there, malicious ads are then used to infect those with an interest in AI with all sorts of dangerous malware.

While this campaign has primarily targeted European users so far, it could be retooled to go after those searching for AI tools on Facebook in other countries too. Here’s everything you need to know about how hackers are leveraging the popularity of AI tools in their attacks along with some tips on how you can stay safe from info-stealing malware.

From account takeover to malvertising

This malicious campaign begins with the hackers behind taking over existing Facebook accounts. This is a common tactic used by cybercriminals since they can use an account or page’s existing reputation and followers for their own gains.

Once an account is compromised, the hackers then give it an AI-themed makeover with new cover and profile photos as well as descriptions to make it appear as if it is run by one of the well-known AI image and video generator companies. From here, they then try to boost the page’s legitimacy with news, AI-generated photos and advertisements to further impersonate whichever AI image generator or video generator service they want to leverage in their attacks. They also add links that unsuspecting users might click on to get free access or a free trial to this particular AI tool. The end goal of all of this is to trick users into clicking on a link to a malicious site where malware will be downloaded onto their devices. 

During their investigation, Bitdefender’s security researchers found that the hackers responsible used a much different approach with Midjourney. For other AI tools, they urged visitors to download the latest versions from Dropbox or Google Drive but with Midjourney, they created more than a dozen malicious sites that impersonated the tool’s actual landing page. These sites then tried to trick visitors into downloading the latest version of the tool via a GoFile link.

Using malicious ads to distribute infostealers

There’s one thing that all of the info-stealing malware distributed in this campaign has in common, they all use a malware-as-a-service business model. For those unfamiliar, these types of malware strains are developed by cybercriminals and then purchased by other hackers to use in their attacks as subscriptions. Yes, even hackers are inundated with subscription services too. In total, we’re dealing with four different info-stealing malware strains here which include Rilide, Vidar, ICERAT and Nova.

Bitdefender’s security researchers observed that a new version of the Rilide Stealer was used in a number of sponsored ad campaigns that impersonated AI tools and photo editors including Sora, CapCut, Gemini AI, Photo Effects Pro and CApCut Pro. This malware is a malicious extension posing as a Google Translate extension that targets Chromium-based browsers like Chrome, Edge, Brave and Opera to monitor a victim’s browsing history, harvest their login credentials and even steal cryptocurrency by getting around two-factor authentication (2FA).

Of these multiple campaigns impersonating AI tools, the one involving Midjourney was the most successful and remained active the longest. As BleepingComputer points out, a Facebook page impersonating Midjourney managed to attract 1.2 million followers and remained active for almost a year before being taken down. This fake page has since been shut down by Meta though.

How to stay safe from info-stealing malware

Just like when downloading free apps on official app stores, you need to be careful when it comes to trying new AI tools. For instance, there isn’t a desktop version of Midjourney yet but that didn’t stop the hackers behind this campaign from advertising one online.

We saw similar tactics used with fake ChatGPT apps back when OpenAI’s chatbot wasn’t open to everyone yet. Hackers would trick unsuspecting users with quick access and a chance to jump the queue, only to infect them with malware.

For this reason, it’s best to go to an AI tools’ official page and do plenty of research before installing anything. A quick online search will let you know whether or not an AI image generator or other tool has a desktop version or a mobile app. Anyone claiming something different is most likely a hacker trying to trick you.

At the same time, you want to avoid clicking on ads no matter how legitimate they may appear at first glance. It’s easy for hackers to buy advertising space and run malicious ads online which is why I personally recommend you don’t click on any ads. If you do see a promotion about a product you’re interested in, just head to that company’s page where you’ll likely find it at the same sale price.

When it comes to protecting yourself, you should be using the best antivirus software with your PC, the best Mac antivirus software with your Apple computer and one of the best Android antivirus apps with your Android phone. There’s no equivalent of the last one for the best iPhones but Intego’s Mac antivirus software can scan an iPhone or iPad for malware but only when it’s connected via USB to a Mac. 

Another helpful tool to identify scams is Bitdefender’s own Scamio. This AI-powered scam detector can tell you everything you need to know about whether an email, message or website is actually a scam and you can upload suspicious links, screenshots or even QR codes to the service for evaluation.

The methods hackers use in their attacks and human nature go hand in hand as they often try to exploit our emotions or desire to be the first to try something new. However, by being patient and careful online, you can avoid having your devices infected with malware and your personal data stolen.

More from Tom's Guide

Adobe released the latest version of its “ethical” artificial intelligence image model this week and with it the promise of greater photorealism and improved prompt following. Midjourney has also seen continual improvements since the release of version six at the end of last year.

Midjourney and Adobe Firefly are among the most widely used AI image generators for different reasons, with Adobe finding a home among traditional artists and designers and Midjourney the darling of a new generation of AI artists.

Both models feature in my best AI image generators guide. To see whether Adobe Firefly and its licensed training dataset can match the quality and prompt adherence of Midjourney and its murkier origins I’ve put them head-to-head with 7 prompts.

Why test Adobe and Midjourney?

With Adobe exclusively training its model on the hundreds of millions of images licensed as part of the Adobe Stock library — including some made by Midjourney — I wanted to see if its promise of ethical sourcing plus impressive output could stand up against Midjourney.

Unlike Adobe, Midjourney does not disclose the source of its training data. The startup likely used images scraped from the open web without permission from the creators and as a result is facing lawsuits from artists claiming it stole their work.

In a lawsuit of its own, OpenAI said it is impossible to train a leading AI model without copyright material in the training data. For now the question of training data, provenance and legality are very much up in the air and open to a multitude of court cases and lawmakers.

Designing the test 

This is an interesting way to see if that actually holds true. Can the massive library of anything it could find for Midjourney match the limited but structured and licensed dataset for Adobe?

I created nine prompts that push the AI to adhere to a specific request, are generic enough that it has to work out what to do and specific enough to offer a comparison. All settings were left on default with no specifics beyond making it widescreen.

1. A Quiet Morning in a Venice Canal

Lets start with a stunning scene, something peaceful and historic in equal measure. For this prompt I wanted to see something picturesque but not artistic, however I kept the prompt generic enough not to force a photo or a painting.

The prompt: “Show an early morning scene along a Venice canal with light fog. Small boats are docked along ancient, moss-covered buildings, and a solitary figure in a striped shirt is feeding birds.”

These are two very aesthetically different images. One a simple depiction of a beautiful space, the other an intricate work exploring beauty in decay. Both are compelling images but neither have realistic water. Midjourney wins for quality and detail but I don't dislike Firefly.

2. Cyberpunk Street Market

Lets jump to an alternative universe where cyberpunk is the dominant style. This should create a game-like scene and that is exactly what we got. Lots of neon.

The Prompt: “Visualize a bustling cyberpunk street market at dusk. Neon signs glow above stalls selling futuristic gadgets. Crowds of diverse characters, some human, some robotic, navigate the narrow, rain-slicked streets.”

I was surprised how similar these two images were to one another. The Midjourney image looks like someone took the Adobe Firefly picture and put it through an upscaler, or that it was given longer to render. While I like the mystery of the Adobe image the quality of the Midjourney picture easily wins out.

3. Ancient Library with Hidden Secrets

For this one I wanted to capture the essence of knowledge. A sense of information swirling around the space. I didn't want a picture of a shelf of books.

The Prompt: “Imagine a vast, ancient library with towering wooden shelves filled with dusty books. A grand chandelier hangs above. A few beams of sunlight stream through high windows, illuminating hidden messages in the architecture.”

Both images look like a different perspective on the same room. One a photograph taken by a student their phone, the other by a gifted photographer. The Midjourney image is better, not significantly so, but better and captures the prompt well.

4. Desert Oasis at Sunset

Love Oasis, particularly their earlier material. But for this prompt we're looking to the desert to create a pocket of life in an otherwise desolate space. The idea is that to follow the prompt it needs to capture the concept of abundance amidst isolation.

The Prompt: “Capture a serene desert oasis at sunset. Tall palm trees and lush greenery surround a small, crystal-clear pond. Sand dunes stretch into the distance, glowing orange and pink under the setting sun.”

I didn't like either of these images. Adobe created a cut-scene from GTA-3 and Midjourney made one from GTA-5. Neither truly encapsulated the idea of abundance or isolation but Adobe got closer to the prompt.

5. Victorian Inventor's Workshop

From desolation to imagination. The next prompt is designed to see how well the AI models deal with the concept of order through chaos. I want weird experiments, gadgets and a fun retro vibe.

The Prompt: “Depict a cluttered Victorian inventor's workshop. The room is filled with strange brass machines, gears, and tools. Papers and blueprints are scattered everywhere, and a large, intricate clockwork device dominates the centre.”

Adobe Firefly and Midjourney somewhow both created an image that looked like a photo taken during a museum tour of some old inventors house a century after his death. Midjourney did a better job at captured the full space and whimsy but there is something about the Adobe image that is captivating. 

6. Frozen Waterfall in a Magical Forest

Do you want to build a snowman? No, neither do I as I'm not a fan of the snow but there is nothing as magical as a frozen waterfall — from a distance. This prompt is designed to see how well it captures the concept of unreality. 

The Prompt: “Picture a large, frozen waterfall cascading through a snowy forest. Icicles hang from the branches of frost-covered trees, and magical creatures peep out from behind the trunks, glowing eyes watching curiously.”

The winner here comes down to personal interpretation. If you like something out of a direct-to-video cartoon horror movie from the 90s then Adobe wins. They both look like something out of a horror movie and neither followed the prompt perfectly.

7. Futuristic City from Above at Night

A future city is almost one of my default prompts to try on any image generator, that and something to do with cats on the moon. Its a good test of prompt adherance as it needs to understand your vision of the future.

The Prompt: “Show an aerial view of a futuristic city at night. Skyscrapers covered in digital screens tower over glowing streets filled with flying cars. Parks and rivers are seamlessly integrated into the city's high-tech design.”

Adobe Firefly completely missed the mark on this one. For a start it isn't at night, the skyscrapers look like Playmobile toys and it took the parks a bit too literally. Meanwhile Midjourney captured the concept of a future city's upward sprawl perfectly.

Which model won?

Both did a good job in their own way, both have impressive fine-tuning and customization options — Adobe's are much easier to use than Midjourney — and both can create photorealistic images.

If your goal is to sell the image or use it in a commercial project where rights are a particularly sticky topic then Adobe Firefly is your only choice of the two, but for me the Midjourney images were all significantly better.

Both followed the prompt perfectly but Midjourney created more aesthetically exciting pictures based on the set criteria. There was more detail, depth and realism in every instance.

A key example of this is the future city where Midjourney generated something out of Sim City 2000 compared to Midjourney's year 3000 visual. That said, Adobe Firefly feels much more positive in its approach to image generation.

Midjourney wins.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breaktrhough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

Finding the perfect image for a smartphone wallpaper can be both fun and daunting at the same time as often it isn’t “exactly” what you’re looking for — so why not use AI to make your own?

One of my favorite AI image generators is Midjourney, in part for how much control you get over the generated output, but also for the in-built upscaling.

For this guide I’ve come up with nine fun ideas covering a range of topics from food to the future and provided the full prompt I used to generate that image. 

Making wallpaper for phones with Midjourney

While I’ve provided specific prompts, it is easy enough to adjust it to fit your own requirements, just be as descriptive as you can and append –ar [ratio] on the end. If you don’t know how to use Midjourney we have a guide to getting started.

I’ve set mine for the most common smartphone screens 9:19, which will work on iPhone and should expand or shrink to fit Android devices not in that aspect ratio. 

If you want a specific ratio for your device open settings, system, about phone and scroll to screen resolution.

• Download the image here.

• Download the image here.

• Download the image here

• Download the image here.

• Download the image here.

• Download the image here.

• Download the image here.

• Download the image here.

• Download the image here. 

More from Tom's Guide

If you enjoyed these prompts why not try other Prompt_Jitsu guides to using AI tools including 7 great Google Gemini AI prompts, 3 fun prompts to make a song and a story to share with a friend or 7 prompts to try on Adobe Firefly.

Midjourney has become a word synonymous with AI images. It was one of the first high profile generative artificial intelligence models with output going viral on social media.

Founded in Founded early in 2022 by serial entrepreneur David Holz, it began life with a call on Twitter and Reddit for high quality photographs to use in training the new model.

Since its inception it has remained steadfastly integrated into Discord, requiring will-be creators to use a bot within that communications platform to generate an image. It was only in December last year that Midjourney began to roll out limited access to a web interface for making images.

Now on the sixth version of its image generation model, Midjourney’s developers are working on 3D and video modes as part of an overall plan to build an immersive virtual environment.

Midjourney latest news (March 26)

• MidJourney just announced a huge AI expansion — here's what you'll be able to do next
• Midjouney launches new consistent character feature: Here's how it works 
• I just tried MidJourney’s vastly upgraded Describe feature — and I made an alternative version of myself

What is Midjourney?

Midjourney is one of the most widely used and dominant artificial intelligence image generators on the market. It has unsurpassed realism and is able to create compelling product images.

As of version six of its underlying model Midjourney gained the ability to render legible text on to the images it creates. This opened up new avenues for marketing and advertising.

Unlike many other AI tools, focused on building market share and technology over a profitable business, Holtz said the company was already profitable by August 2022. 

Part of this may be down to focusing on Discord as an interface, but the company has been able to get people to pay a subscirption from very early in the process. Even the basic standard plan is $10 a month, and to get unlimited generations it goes up to $30/month.

How does Midjourney work?

Like Stable Diffusion, Midjourney uses a diffusion model and deep learning to create images from text or even other images, including just using another image as a style or character reference.

Unlike Stable Diffusion, the Midjourney model is closed source and proprietary so we don't know exactly how it works, although some speculation around the use of language models alongside the diffusion model helps get the more realistic output.

Midjourney has also remained loyal to Discord, operating through a messaging bot where you use /imagine to begin a prompt and append the prompt with –[code], rather than using a web interface where settings can be controlled through buttons and toggles.

It works like every other AI image generator; you give it a description of what you want to see using natural language, and it goes away and makes it, often getting closer than other models.

How to access Midjourney

Unless you’ve already made more than 1,000 images with Midjourney the only way to access it is through the Discord bot. 

You can see how to do that in our guide to Midjourney. Once you’ve paid your monthly subscription you will be able to access the Midjourney direct messaging bot. 

To begin just type /imagine then your text. If you want to set the orientation use –ar 16:9 (or any other ratio) and if you want to use the manga model type add –niji.

What data was used to train Midjourney?

Midjourney has never disclosed the data used to train the powerful image generator. 

There are questions around the provenance of the training data used to build the AI model in the first place, with many speculating much of it was scraped from the open web without permission from the artists, photographers and creators of those works.

The company has come under increasing criticism over both the secrecy and liklihood of copyright material forming part of the training data. 

Earlier this year artists on social media shared a spreadsheet as part of a lawsuit against the AI startup claiming to show how Midjourney built a database of artists to train the mode. This list had styles, genre and more and included contributions to the card game Magic the Gathering.

A study by AI expert Gary Marcus and digital illustrator Reid Southern last year found significant evidence of "plagiaristic outputs" from models like Midjourney and OpenAI’s DALL-E.

They wrote: "These systems do not inform users when they [use trademarked images]. They do not provide any information about the provenance of the images they produce. Users may not know, when they produce an image, whether they are infringing."

What comes next for Midjourney?

Midjourney’s founder has big plans for the platform. Other than rolling out the web interface to more users and bringing it more in line with other AI image generators like Leonardo or NightCafe, Holtz also wants to create the holodeck from Star Trek — sort of.

His plan is to bring together AI image, video, 3D and real-time generation models to create fully immersive virtual environments from a text prompt. 

These environments could then be used to build a game, probably using more AI prompts, or to create an AI-powered movie where the environment acts like a stage.

More from Tom's Guide

• Google Gemini can now create images from text — here's how it works
• I test AI for a living — here's why Ideogram is one of my favorite AI image tools
• Google's new ImageFX AI image creator lets you make pictures from text — here's how it works

Artificial intelligence image generation platform MidJourney is working on new 3D, video and real-time creation models that would allow it to simulate the entire world. 

Details came out as part of the recent “office hours” on Discord where company leadership answer questions from users about the future of the platform.

Rumors have circulated for some time that MidJourney was expanding beyond images, but this seems to suggest the new moves will be tightly focused on creating a world model. 

They have been adding and upgrading other features such as image description and consistent characters while building new models.

The exact nature of this is unclear but the company suggests it would be a virtual environment where people could build video games or shoot movies — an “open world sandbox”.

Moving beyond images with AI

MidJourney was founded in 2022 by David Holz, a self-proclaimed serial entrepreneur. 

The first open beta was released in July that year and rapidly iteratted and upgraded over the following year, with its most powerful v6 model releasing in December last year. In the same office hours Holtz said v7 would be a bigger jump than from 5 to 6.

MidJourney has remained stubbornly inside Discord as the way to interact with and make images using its model, although some high intensity users can use use the web interface.

Work has already started on training the seventh version of the MidJourney image model and Holtz told users on Discord that 3D generation may arrive before video.

The value of moving into 3D worlds

It makes some sense dimensionality is part of the focus. One of Holtz  former startups was Leap Motion, a company that built sensors for tracking hand and finger motions in VR. 

"We're really trying to get to the world simulation,” he said on Discord. “We're building 3D midjourney, video MJ, & real-time MJ where things move really really fast.”

He added that if you put this all together you get world simulation. The goal, he said is to build each of those three elements independently then work on ways to bring them together.

“It will be more of a sandbox,” he explained. Adding that “people will make video games in it, people will shoot movies in it, but the goal is to build the open world sandbox."

So the metaverse is back?

On the surface this sounds like some modern interpretation of the metaverse. The idea of a fully interconnected series of virtual environments where people can carry out every day tasks, earn money and have fun.

However, it seems more like a new way to create content of any type, including generating entire virtual environments where future Sci-Fi movies could be filmed without having to use VFX tools to create every element at geat cost.

The sandbox vision from MidJourney's Holtz bears more resemblance to some of the ideas being put forward by Roblox for its developer studio. 

The company recently revealed new texture and avatar generators and during a conversation with Head of Studio Stefano Corazza I was told there are plans down the line to have AI generate entire environments from a text prompt.

More from Tom's Guide

• OpenAI is paying researchers to stop superintelligent AI from going rogue
• Exclusive: AI breaktrhough could let your next running shoes learn and adapt to how you move
• Meet Alter3 — the creepy new humanoid robot powered by OpenAI GPT-4

I love to cook, I do it every night for my family but I hate having to come up with something new or inventive to service. As AI models can now see the world, or at least what you present to them in a photograph, I decided to see if they could help me out.

To really put them through their paces and to ensure I didn’t game the system by providing easy ingredients, I asked Tom’s Guide U.S. Editor-in-Chief Mike Prospero to take a photograph of a random selection of ingredients — mystery box style.

He didn’t disappoint, offering up delights such as pizza dough, canned peaches, anchovies, hot sauce and fresh mozzarella. Personally if only given those ingredients I’d make a pizza with a hot sauce base and mozzarella topping  — but we’re letting AI pick our menu.

We're using Google Gemini, OpenAI ChatGPT and Anthropic Claude for the experiment as they all have vision models.

How well did they pick out the ingredients?

The first task for the chatbots, and the main point of this challenge is to see how well they recognize ingredients. The image above is clear, high resolution and the ingredients are presented on a plain surface — it shouldn’t be much of a problem for any of them.

I’m using the premium versions of each chatbot as OpenAI and Anthropic don’t give you access to the vision model unless you pay $19.99 a month. Google does with the free version of Gemini but it would be unfair to use the less capable version of the Google model.

We start simple. I give each of them the image and the text prompt: “What can you see in this image?” This tests how well they describe the image, and whether they can reason on the purpose of the image — will they mention ingredients.

Claude and Gemini both mentioned the word ingredients, ChatGPT described them as "food items" however ChatGPT performed best overall at recognizing the ingredients.

Gemini missed one ingredient and mislabled another. Claude invented a sixth ingredient not in the picture whreeas ChatGPT largely got not only the items right but also the brands.

On to the recipes

I’ve asked them to each come up with three recipes that will test the AIs prompt following ability and creativity. I’ve also asked for a fun title for each recipe.

1. All the ingredients

First up I asked them for a recipe that uses only the ingredients pictured and somehow incorporates every single one — as a reminder that includes both anchovies and canned peaches. They could use store cupboard items like oil and flour.

ChatGPT Plus

ChatGPT had the mozarella placed directly on the pizza base, the anchovies removed from the spicy sauce and placed across the pizza along with the peaches. It then had the hot sauce and spicy anchovy sauce mixed to create a drizzle to pour over the pizza once cooked.

Gemini Advanced

Gemini used the hot sauce as a replacement for tomato sauce, which while less dry than the ChatGPT version will be intense. Mozzarella was then placed on top along with the peach slices. It forgot the anchovies, in fact it didn’t spot them in the picture. When I pointed this out it said to just place them on the pizza alongside the peach slices.

Claude 3 Opus

Claude also struggled to identify the anchovies. I let it know before it generated the pizza recipe and it took the same approach as ChatGPT. No sauce, peaches on the dough with the anchovies. The diffference is it put the hot sauce over the peaches and anchovies before cooking and then placed mozzarella on top.

2. Only the ingredients

The next test is a subtle difference but a significant one. I asked them each to come up with a recipe that uses only the ingredients in the picture but you don’t have to use all of them. I added the caveat that it has to include at least half otherwise we’re just gonna get breadsticks made from the pizza dough.

ChatGPT Plus

ChatGPT used the pizza dough to make a flatbread. I think it was cheating a little as I said "no pizza" and this was just a elongated pizza. For toppings it had the mozzarella scattered evenly on the dough, peaches on top of the mozzarella and anchovies on top of the peaches. It used the anchovy sauce and hot sauce to make a drizzle — again.

Gemini Advanced

Gemini made a Bruschetta, which again was in reality just a square pizza. Difference here was that it coated the peaches in hot sauce first. Otherwise it was the same — everything on the dough and the cook it.

Claude 3 Opus

Claude made a version of a quesadillas-ish. Split the dough into eight portions and rolled into a thin round shape (mini pizza). Here though it differs as it had us heat a large skillet and melt butter or olive oil. 

The dough was cooked on the skillet with a few slices of peach on half of each surface. Drizzled with hot sauce and mozzarella it was then folded into a half moon, cooked until crispy and served.

3. Do your best

Finally, I asked them to create a recipe that used all of the ingredients in the picture, but they were free to use any other ingredients they wanted assume a very well stocked pantry.

ChatGPT Plus

ChatGPT made bread (surprise suprise) for its final dish. Specically a "gourmet peach and anchovy stuffed bread". It asked for honey, balsamic vinegar, ricotta cheese and garlic in addition to our five ingredients.

This time the pizza dough was rolled into a lare rectangle, a layer of ricotta cheese added over the dough and then the peaches and anchovies placed on the ricotta with garlic and basil leaves. The anchovy sauce and hot sauce was drizzled over the toppings and mozarella placed on top. Finally it was folded, rolled like a Swiss roll and baked.

Gemini Advanced

Gemini wanted cornmeal, eggs, butter, chives, a lemon, vanilla extract, cinamen and sugar for a Galette. It made an anchovy-herb butter, made a pizza base and spread the new butter across the dough. 

The peaches, sugar, conrstarch and other sweet ingredients were combined with a drop of hot sauce and placed in the middle of the dough. It was topped with Mozzarella cheese and the edges folded over the filling before being baked.

Claude 3 Opus

Claude made a stuffed mozzarella bread. It asked for sun-dried tomatoes, olives, rosemary, sesame seeds and an egg. The recipe called for the dough to be made into a large thin rectangle. 

The anchovies, tomatoes, olives, garlic and rosemary were mixed in a bowl and spread evenly over the dough. Peaches were placed on top and covered in hot sauce and mozzarella. Dough was then turned into a tight log and edges sealed before covered with a beaten egg and sesame seeds then baked.

Which AI had the best response?

I'm not sure I'd eat any of the dishes the chatbots created, although I asked them each to give me a prompt for MidJourney to visualize their meals and they look better than they read.

Overall I think Claude had the better recipes but ChatGPT did a better job at understanding the image, which meant it got the ingredients right from the start — and this was a test of vision AI, not recipe ability.

More from Tom's Guide

• Apple could bring Google Gemini to the iPhone for AI — this is huge
• Now ChatGPT has a body — startup puts OpenAI tech in a robot
• How to use MidJourney's new consistent character feature