Those who have spent any time with AI video generators, know just how frustrating it can be when what you've prompted and what you've generated are completlely different. It's frustrating when it comes time to edit those videos, especially when you've generated something impressive, but need to tweak a single thing such as removing a person or changing the lighting.
Adobe is trying to fix that.
Today, Adobe is rolling out a set of video-focused features for Firefly designed to make AI video editing feel more like real editing — iterative, controlled and integrated into a broader workflow. The updates include prompt-based video edits, camera motion references, a browser-based Firefly video editor and built-in upscaling powered by a partner model.
Here's what's new
Users can now prompt to edit their videos. Instead of regenerating an entire clip every time you want a change, Firefly now lets you edit an existing AI-generated video using text prompts. That means you can do things like:
- Remove or replace objects or people
- Change the background
- Adjust lighting, contrast or atmosphere
- Reframe or zoom without starting over
This is a big shift. Most AI video tools still treat every change as a full regeneration, which makes fine-tuning nearly impossible. Adobe’s approach is closer to how creators already think: generate first, then refine.
Camera motion reference keeps shots consistent
Firefly’s Video Model now supports camera motion reference, which solves another common AI video problem: inconsistent movement.
You can upload a start-frame image (your scene) and a reference video that shows the camera movement you want. From there, Firefly then applies that camera motion — pan, dolly, zoom, tilt — to your generated clip while keeping the subject and framing anchored.
This makes it far easier to create cinematic-looking shots that don’t feel jittery or random, especially for B-roll, transitions or short-form video.
Built-in video upscaling to 4K
Adobe is also integrating Topaz Astra, a partner AI model focused on video enhancement and upscaling. With this integration, users can now enhance low-resolution or older footage, upscale clips to 1080p or 4K and queue multiple upscales while continuing to work.
This is aimed at creators who already have footage (not just AI-generated clips) and want a quick way to improve quality without leaving Adobe’s ecosystem.
A new Firefly video editor (public beta)
Firefly is also getting a browser-based video editor, now available in public beta.
It supports two workflows, including timeline editing, with multitrack control for video and audio, and text-based editing, where you edit video by working with a transcript (ideal for interviews, explainers, or talking-head content)
The editor lets users combine their own footage with Firefly-generated video, plus music and audio.
Bottom line
Adobe’s long-term vision is clear: AI shouldn't replace human editors, it should work with them. By focusing on prompt-based edits, camera consistency and practical tools like upscaling and transcript editing, Adobe is positioning Firefly as a creative hub rather than a novelty generator.
To encourage experimentation, Adobe is also offering a limited-time unlimited generations promotion (through January 15) for certain paid Firefly plans, including Pro, Premium and high-credit tiers.
The promo covers Firefly image models, Firefly video models and select partner image models. Unfortunately, it doesn’t apply to free users, but it does lower the friction for creators who want to seriously test Firefly’s new video tools without worrying about credits.
More from Tom's Guide
- GPT‑5.2 is way smarter than I expected — these 9 prompts prove it
- I'm a former Disney employee — here’s why the $1B deal with OpenAI is different than people think
- 9 real mistakes people make with ChatGPT — and how GPT-5.2 fixes them
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.
You must confirm your public display name before commenting
Please logout and then login again, you will then be prompted to enter your display name.
