I asked Claude to write image prompts — then tested ChatGPT vs Nano Banana to crown a winner

AI image generators are getting shockingly good — but the results you get often depend on one thing: the prompt.

Instead of writing the prompts myself, this time, I decided to try something unusual. I asked Claude, Anthropic’s chatbot, to generate a series of highly detailed prompts covering everything from cinematic sci-fi scenes to surreal abstract art.

While the faceoff was between ChatGPT and Gemini's Nano Banana, this also tested Claude's mutlimodal ability. Even though Claude doesn't have the ability to generate images, this test underscores it's ability to describe images in great detail.

So the prompts it created were fed directly into ChatGPT-5.4, which has newly updated its image capabilities, and Nano Banana 2. These models are two of the most capable AI image tools available right now. The idea was simple: if both models used the exact same prompt, which one would produce the stronger image?

To find out, I ran the seven tests spanning photorealism, fantasy, cinematic scenes, abstract art and nature photography. Some results were close. Others were surprisingly lopsided.

Here’s what happened when Claude became the secret prompt engineer.

1. Photorealistic

Claude’s prompt: "A weathered fisherman in his late 60s sitting alone on a fog-covered dock at 5:47 AM. His hands are calloused and scarred, wrapped around a chipped ceramic mug of black coffee. He wears a faded olive oilskin jacket with a torn breast pocket. The dock planks are dark with moisture, reflecting pale amber dock lights. A small wooden fishing boat bobs gently behind him, its hull painted in peeling cobalt blue. Shot on a Canon EOS R5, 85mm f/1.4 lens, golden hour diffused through marine fog. Hyperrealistic, 8K, photojournalistic."

ChatGPT created a tight, intimate portrait. The focus is heavily on the man's face and his emotional state (somber/reflective). The image offers more emphasis on the grime and wear of the clothes and skin. It feels a bit more "painterly."

Nano Banana generated a wide environmental shot. It captured the scale of the dock, the full length of the boat and the perspective of the pier with higher photographic clarity.

Winner: Nano Banana wins for the more photojournalistic image because it feels slightly more authentic with a wide-angle perspective and the way the wet dock reflects the light.

2. Fantasy

Claude’s prompt: "A colossal ancient library built inside a hollowed-out living tree, its interior stretching hundreds of feet upward. Thousands of glowing amber lanterns hang from spiraling staircases carved into the bark. Books float autonomously through the air, guided by invisible currents of golden magic. Tiny scholars in deep jewel-toned robes walk suspended bridges. Roots pierce through the floor and ceiling, thick as pillars. Rendered in the style of a detailed oil painting — rich chiaroscuro lighting, impasto brushwork, inspired by Rembrandt and Alan Lee. Warm amber and deep forest green palette."

ChatGPT leaned way into Rembrandt-style chiaroscuro. It is much darker, moodier and uses extreme contrast between light and shadow.

Nano Banana gave a wild, panoramic style image that was much brighter and far more detailed.

Winner: Nano Banana wins for what feels like a truly illustrative masterpiece.

3. Astronaut

Claude’s prompt: "A lone astronaut in a next-generation matte-white EVA suit stands on the fractured ice surface of Europa, Jupiter's moon. Jupiter dominates 40% of the sky behind her — banded in burnt sienna, cream, and storm-grey, its Great Red Spot barely visible at the horizon. The ice beneath her feet is cracked into geometric plates, glowing faint blue from subsurface bioluminescence below. Her helmet visor reflects the gas giant in perfect convex detail. Cinematic anamorphic lens, 2.39:1 aspect ratio, Ridley Scott-esque lighting, volumetric atmosphere, IMAX-quality render."

ChatGPT focused on the "grandeur" and the journey. We see what the astronaut sees, making the scale of Jupiter feel more overwhelming.

Nano Banana focused on the subject. The astronaut is facing the camera, which emphasizes the gear and the reflection in the visor.

Winner: Nano Banana wins here for adhering to the technical camera specs (anamorphic lens and 2.39:1 ratio) which makes it look like a genuine still from a modern sci-fi film.

4. Abstract

Claude's prompt: A human heart made entirely of blooming red peonies and dark thorned vines, suspended mid-air in a vast black void. From its chambers, golden clockwork mechanisms spill outward — tiny gears, springs, and escapements drifting in zero gravity. Droplets of deep crimson and liquid gold hang frozen in the air around it. The composition is symmetrical but imperfect — organic chaos meeting mechanical precision. Style: dark surrealist oil painting meets botanical illustration, reminiscent of Alex Grey and Ernst Haeckel. Ultra-detailed, baroque composition.

ChatGPT generated flowers and vines that are deeply integrated, looking as if they are the heart structure itself and growing together.

Nano Banana created a heart form that acts more like a cage, and the peonies are attached to or blooming out of it.

Winner: ChatGPT wins for a bright, detailed and stylized gold filigree that aligns well with a "Baroque" aesthetic.

5. Nature

Claude's prompt: "A massive thunderstorm cell over the American Great Plains at golden hour. The supercell's rotating wall cloud descends toward flat golden wheat fields below. The sky is split — violent deep purple and charcoal on the left, blazing amber and pink on the right where sunlight still breaks through. Lightning strikes three points on the horizon simultaneously, their bolt structures fully detailed and branching. A lone red barn sits tiny in the foreground. Shot with a wide-angle 16mm lens, HDR, ultra-high detail in the cloud structures, meteorologically accurate cumulonimbus formation."

ChatGPT delivered a highly dramatized image with an intense fiery sky that gives it almost an apocalyptic or surreal feeling.

Nano Banana generated an image that feels more like a real photograph with
subtle transitions and realistic cloud textures.

Winner: Nano Banana wins for technical realism that feels more like a high-end photography print.

6. Samurai

Claude's prompt: "A Japanese feudal-era samurai standing still in the center of a burning village at midnight. His black lacquered armor is cracked, one shoulder plate hanging loose. Cherry blossom petals mix with falling ash in the firelit air. His katana is drawn and gleaming, reflecting the orange inferno around him. His expression is unreadable — neither victorious nor defeated. Behind him, the silhouette of a wooden gate collapses in flame. Cinematic lighting, dramatic low angle shot, inspired by Akira Kurosawa's visual language but rendered with modern photorealism. Deep shadows, extreme contrast, 4K film grain."

ChatGPT delivered a tight, vertical framing image that included a massive samurai looming and dominating the entire frame.

Nano Banana seems to prefer wide, cinematic shots. This one captured the scale of the village and the burning torii gate, providing more environmental context.

Winner: ChatGPT wins for a dramatic character that almost pops out of the screen to tell a story.

7. Whimsy

Claude's prompt: "A tiny independent bookshop on a rainy London side street at dusk, seen from across the cobblestone road. Warm golden light pours from its fogged windows. A handwritten chalkboard sign leans against the door. The shop's facade is painted deep forest green with gold lettering. Potted lavender flanks the entrance, petals jeweled with rain droplets. A tabby cat sits in the window between stacked novels. Reflections of lamplight shimmer on the wet cobblestones. Style: painterly realism — loose impressionistic brushwork with sharp focal detail, inspired by Edward Hopper and Childe Hassam. Cozy, melancholic, intimate."

ChatGPT generated an image that feels like a passerby glancing at the window and the generic “Bookshop” above the door feels like an afterthought.

Nano Banana added special details, including a unique name for the bookstore, gentle lighting and books with spines that are actually visible.

Winner: Nano Banana wins for extra creativity with “The Dusty Spine” naming and a level of personality that seems to fit the prompt much better.

Final thoughts

After running seven different prompt tests, it's clear Nano Banana consistently excels at environmental scale and photographic realism, often producing wider compositions that feel like real camera shots. That advantage helped it dominate categories like photorealism, nature and cinematic landscapes.

ChatGPT, on the other hand, tends to prioritize storytelling and subject focus. Its images frequently center on characters and emotional detail, which worked especially well for the abstract and samurai prompts.

What makes this experiment interesting is that neither model had a prompt advantage. Claude wrote every prompt, meaning both generators started with the exact same creative instructions. The results suggest something important about AI image tools right now: the difference isn’t just the prompt, but how each model interprets it.

And as AI models continue evolving, the real power might come from combining them, letting one model design the prompt while another generates the image. In other words, the future of AI creativity may be less about choosing one tool — and more about orchestrating several of them together.

Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds.

More from Tom's Guide

• I use this ChatGPT ‘alpha prompt’ to make stronger arguments — here’s the exact prompt
• I can text Claude like a personal assistant now — this iPhone shortcut made it possible
• I built a library of 'thinking prompts' for Claude — these are the ones I use most