ByteDance launched Seedance 2.0, its latest generative video model, last Wednesday through its Dreamina AI platform and social media has been abuzz with its creations. Within 24 hours of the release, a two-line text prompt had produced a photorealistic video of Tom Cruise fighting Brad Pitt on a rooftop. The Motion Picture Association, SAG-AFTRA, Disney, and Paramount have all now publicly condemned ByteDance, with the studios sending formal cease-and-desist letters demanding it stop.
The model accepts text, image, audio and video inputs and generates clips up to 15 seconds long. Unlike previous AI video tools, Seedance 2.0 combines synchronised audio and video with what ByteDance calls "director-level controls" over camera movement, lighting and character consistency. The results look close enough to real studio footage to prompt Deadpool screenwriter Rhett Reese to post, 'I hate to say it. It’s likely over for us.’
What went down
ByteDance launched Seedance 2.0 on 12 February via its Dreamina AI platform.
Viral deepfakes of Hollywood IP appeared within hours of the launch.
The MPA condemned ByteDance for copyright infringement on a "massive scale."
Disney sent a cease-and-desist alleging a "smash-and-grab" of its IP.
Paramount followed with its own legal demand citing South Park and Star Trek.
Why does this matter?
The output of Seedance 2.0 appears to be really, really ‘good’. As in, from one prompt you can now create detailed and realistic action scenes. Early trialists and testers on social media have been astonished at it’s capability. Creating entertaining and personalised content from a prompt is now on the horizon.
The copyright issues this causes are not so ‘good’. OpenAI's Sora 2 triggered similar concerns in October 2025. The difference this time is the pace: cease-and-desist letters arrived in under 72 hours, suggesting studios have moved from reactive alarm to prepared legal strategy. The speed of the backlash shows how seriously the industry is treating generative AI video.
Our take
Veo 3 caused a good helping of astonishment when it was released last May. However, after putting it through a rigorous practical test in September, we deemed that it still wasn’t really production ready. Generating realistic video for anything that required accuracy (like dinosaurs) was still frustratingly out of reach.
This time will likely be different. It’s been almost a year since Veo 3, and 3 months since Sora 2 was released. If the outputs of Seedance 2.0 are as production ready and accurate as they appear to be, this will cause a surge in usage and transform how video content is made.
But the legal implications are also significant. It’s familiar territory. A Chinese AI company releases a model with minimal guardrails. The output goes viral. Hollywood responds with legal threats. ByteDance claims to have disabled the ability to generate clips of real people.
Seedance 2.0 will be available through ByteDance's existing consumer apps, meaning the audience is already built in. When the global rollout via CapCut arrives, the potential volume of AI-generated content using copyrighted material will be orders of magnitude larger. Studios are preparing for that moment now. The question is whether their legal frameworks can keep pace with models that are improving every quarter.
Another big thing… AI safety in question
It was an extraordinary week for AI safety. On Monday 9 February Daisy McGregor, Anthropic’s Head of UK Policy agreed that Claude was ready to kill someone when researchers told it that it would be shutdown. The same day Mrinank Sharma, who led Anthropic's Safeguards Research team, resigned and published an open letter warning that "the world is in peril." Zoë Hitzig, a researcher at OpenAI, also announced her resignation in a New York Times op-ed citing concerns about the company's advertising strategy.
