Adobe is moving Firefly into video, but its first step is not only about creating clips from a prompt. The company is also using generative AI to solve a familiar editing problem: footage that ends just before the editor has the beat they need.
The launch brings video generation capabilities to Firefly ahead of Adobe MAX, while a related feature, Generative Extend, arrives in the Premiere Pro beta app. Together, the tools show how Adobe is trying to make AI video useful to creative professionals without asking them to abandon the workflows they already use.
What Adobe Firefly can now generate
Users in a limited beta can sign up on Adobe's website to test Firefly's video generator. The web beta includes both a text-to-video model and an image-to-video model, with each producing up to five seconds of AI-generated video.
The web beta is free to use, though it likely has rate limits. Adobe says Firefly has been trained to create both animated content and photo-realistic media, depending on what a user asks for in a prompt.
The Firefly video web app also includes controls that matter to people thinking visually. Users can adjust camera pans, the intensity of camera movement, angle, and shot size. Those settings point to a product aimed at directors, editors, designers, and other creators who need more than a single generic output from an AI video generator.
Adobe also says Firefly is capable of producing videos with text, at least in theory. That is notable because AI image generators have historically struggled with text, making it a practical test of how reliable the system can be in real creative work.
Why Generative Extend may matter more
The more immediately useful feature may be inside the Premiere Pro beta app. Generative Extend uses Firefly to lengthen video clips by up to two seconds, creating a short continuation of a scene.
The feature is designed to add an extra beat while continuing camera motion and the subject's movement. For an editor, that can mean having just enough room for a transition, a pause, or a cleaner rhythm in a sequence.
Generative Extend also extends background audio. Adobe says this gives the public its first taste of the AI audio model the company has been quietly working on. The audio extension has an important limitation: it will not recreate voices or music, a choice meant to avoid copyright lawsuits from record labels.
In demos shared with TechCrunch ahead of the launch, Generative Extend produced more impressive results than Firefly's text-to-video model and appeared more practical. The text-to-video and image-to-video tools did not seem to have the same polish or impact as competing AI video systems such as Runway's Gen-3 Alpha or OpenAI's Sora, though Sora has yet to ship.
Adobe is betting on AI editing first
Adobe appears to be prioritizing AI editing features over pure AI video generation. That choice fits the company's existing creative audience, which includes users who expect exact control over pixels, frames, and finished assets.
Adobe's VP of generative AI, Alexandru Costin, described that audience to TechCrunch as highly demanding. He said, "Our audience is the most pixel perfect audience on Earth," and added that users want AI to help extend, vary, or edit existing assets more than they want it to create new ones from scratch.
That distinction matters. A blank-canvas AI video model may be attention-grabbing, but a tool that fixes a short clip can fit directly into an editor's day. Adobe's first Firefly video feature is therefore aimed at a clear production pain point rather than a broad promise of automated creation.
The strategy echoes Adobe's earlier success with Firefly's image model in Photoshop. Adobe executives previously said Photoshop's Generative Fill feature is one of the most used new features of the last decade, largely because it complements existing workflows and speeds them up.
The creative tension around AI video
Adobe is launching these tools into a creative community that has mixed feelings about generative AI. The company is trying to compete in a crowded field of AI startups and technology companies showing increasingly capable models. At the same time, many creatives worry that AI features could replace work they have spent decades doing with a mouse, keyboard, and stylus.
Adobe has tried to address that concern in part through how it trains Firefly. The company has reportedly paid photographers and artists $3 for every minute of video they submit to train the Firefly AI model. Even so, the source article makes clear that many creatives remain wary of AI tools or fear being made obsolete by them.
Costin argues that generative AI will increase demand for creative work rather than reduce it. He pointed to companies wanting individualized and hyper personalized content for users, describing that demand as effectively unlimited.
He also framed AI as part of a broader technological shift for creative professionals, comparing it with digital publishing and digital photography. His message to hesitant creators was direct: "Take advantage of generative capabilities to uplevel, upskill, and become a creative professional that can create 100 times more content using these tools," he said. "The need of content is there, now you can do it without sacrificing your life. Embrace the tech. This is the new digital literacy."
Watermarks and commercial safety
Firefly-generated videos will automatically include "AI-generated" watermarks in their metadata. Meta uses identification tools on Instagram and Facebook to label media with these labels as AI-generated, which means platforms or individuals may be able to identify media as synthetic when the relevant metadata is present.
There is an important visibility issue, however. Adobe's videos will not include default visible labels that clearly tell human viewers they are AI-generated. The identification is built into metadata, not necessarily into the picture a person sees on screen.
Adobe also describes Firefly as designed to generate "commercially safe" media. The company says it did not train Firefly on images and videos that include drugs, nudity, violence, political figures, or copyrighted materials. In theory, that should limit the creation of "unsafe" videos.
The real test begins as more people use the model. With free access to Firefly's video model through the limited beta, Adobe now has to prove that its AI video tools can be useful, controlled, and acceptable to the creative professionals it most needs to win over.