ByteDance’s Seedance 2.5 is now blocking roughly 80% of prompts asking it to generate copyrighted Hollywood characters and footage, according to research the AI-copyright testing startup VN.ai shared exclusively with Deadline. That’s the company’s answer, six months late by industry standards, to the mess its previous model made when it let a two-line prompt turn Tom Cruise and Brad Pitt into rooftop brawlers who never actually fought.
What Seedance 2.0 Actually Did
When Seedance 2.0 launched, users found within a day that it would generate near-photoreal video of real actors doing things they never did, on top of unauthorized riffs on Avengers: Endgame, Game of Thrones, Titanic, Lord of the Rings, and Friends, among others. The Cruise-Pitt fight clip was the one that actually went viral, and it triggered a response fast even by AI-controversy standards: the Human Artistry Campaign, whose members include SAG-AFTRA and the Directors Guild of America, called the launch “an attack on every creator around the world,” adding that “stealing human creators’ work in an attempt to replace them with AI-generated slop is destructive to our culture.” SAG-AFTRA separately said it “stands with the studios in condemning the blatant infringement enabled by ByteDance’s new AI video model Seedance 2.0.” Disney sent ByteDance a cease-and-desist letter over the same launch, a separate matter from the still-unresolved Disney, Universal, and Warner Bros. Discovery lawsuit against the AI company MiniMax over its Hailuo platform.
How the New Filter Actually Works
The technical approach VN.ai describes is more interesting than a simple keyword blocklist. Seedance 2.5 generates the video first, then moderates the finished clip before it ever reaches the user, rather than screening the prompt text up front. That only works if the model was trained on enough copyrighted material to recognize what its own output resembles, which is its own uncomfortable admission buried inside a fix. VN.ai’s testing found the new version blocks around 80% of infringing prompts, a level of consistency the researchers called “far more” reliable than Seedance 2.0’s essentially nonexistent guardrails. A ByteDance spokesperson told Deadline the company “launches its models with AI transparency measures and guardrails intended to prevent unauthorized use of intellectual property,” and that it “continues working to strengthen those protections alongside creative industry partners.”
80% Still Means One in Five Gets Through
The gap in that number is the real story. VN.ai’s researchers found workarounds that still slip past the filter, including generating a recognizable DreamWorks-style character without naming the studio or franchise directly, which is exactly the kind of semantic dodge that makes automated moderation an arms race rather than a solved problem. None of the underlying legal disputes have moved either. The Motion Picture Association’s core objection was never really about output filtering; it was about training-data infringement happening before a single prompt gets typed, and a better output filter doesn’t touch that argument at all.
Why Editors Should Care Beyond the Lawsuit
This isn’t an abstract legal fight for anyone cutting video professionally. Every editor now weighing an AI-generated placeholder shot or B-roll fill is inheriting the same provenance risk Seedance 2.0 made undeniable: a clean-looking generated clip can still be built on someone else’s copyrighted performance or footage, invisibly, with no way for the person pasting it into a timeline to know. An 80% block rate is progress, but it’s also proof that the industry’s current answer to “is this generated clip actually clear to use” is still a probability, not a guarantee.
ByteDance fixing four out of five prompts is a real improvement over fixing essentially none. It’s also a tacit admission that the fifth one is still out there, waiting for whoever finds the right phrasing. Until studios, unions, and AI companies settle what training on copyrighted film actually requires in compensation or consent, every generative video tool an editor reaches for carries that same asterisk.
Sources:
- TheMovieBlog — “Seedance 2.5 AI Video Generator Now Blocking Hollywood Intellectual Property” (citing Deadline/VN.ai research)
- Comic Basics — “After Cruise-Pitt Deepfake Fallout, ByteDance’s Seedance 2.5 Is Now Blocking Hollywood Rip-Offs”
- Deadline — “Seedance 2.0 Hollywood Deepfakes Slammed As ‘Destructive To Culture’ By Human Artistry Campaign”
- TechCrunch — “Hollywood Isn’t Happy About the New Seedance 2.0 Video Generator”
- Axios — “Scoop: Disney Sends Cease and Desist Letter to ByteDance Over Seedance 2.0”
