Six weeks ago, a video of Tom Cruise fighting Brad Pitt on a rooftop went viral. It was, of course, not real. It was generated by Seedance 2.0, ByteDance’s AI video model, and it set off a firestorm that drew cease-and-desist letters from six major Hollywood studios, a formal denunciation from the Motion Picture Association, and a pointed rebuke from SAG-AFTRA over the unauthorised use of its members’ likenesses. Rhett Reese, the screenwriter behind the Deadpool films, watched the clip and offered a blunt assessment of the technology’s implications for his profession.

Now ByteDance is attempting something delicate: relaunching the very tool that provoked that backlash, but with enough safeguards to make the case that it has heard the criticism. On Wednesday, the TikTok parent company said its global safety and intellectual property teams had worked with a third-party red-teaming partner to bolster Seedance 2.0 ahead of its international release through CapCut, ByteDance’s video editing platform, which reports more than 400 million monthly active users.

The new safeguards are substantive, at least on paper. Seedance 2.0 now blocks video generation from images or videos containing real faces, a direct response to the deepfake controversy that engulfed the model in February. CapCut will also block the unauthorised generation of copyrighted characters, addressing the parade of AI-rendered Shreks, SpongeBobs, Darth Vaders, and Deadpools that the MPA had cited in its complaint.

On the transparency front, all output will carry both visible watermarks and embedded C2PA Content Credentials, the industry-standard protocol for identifying AI-generated content across platforms. ByteDance is also introducing what it calls an “advanced invisible watermarking” technology designed to identify content made with the model even after it has been shared or altered off-platform, and the company says it will conduct proactive monitoring for IP violations.

The rollout itself reflects a calculated caution. CapCut will initially make Seedance 2.0 available to paid users in Brazil, Indonesia, Malaysia, Mexico, the Philippines, Thailand, and Vietnam. Conspicuously absent from the list are the United States and India, ByteDance’s two most complex regulatory markets. Europe, Africa, South America, and Southeast Asia are expected to follow, according to the company, though no firm timeline has been offered for the US.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

The AI video arms race

The timing of the relaunch is notable. Just days earlier, OpenAI announced it was shutting down Sora, its own AI video generation tool, after downloads fell 45 per cent by January and a licensing deal with Disney collapsed. Where OpenAI retreated, ByteDance is advancing, though into a market now acutely sensitised to the regulatory questions that AI-generated content raises.

The EU AI Act’s transparency requirements, which take effect in August 2026, will mandate that providers of generative AI systems mark their output in machine-readable formats and disclose the artificial origin of deepfakes. ByteDance’s adoption of C2PA watermarking and invisible marking appears to anticipate these obligations, though whether its safeguards will satisfy European regulators remains to be seen.

Red-teaming reports suggest the guardrails are not impenetrable. According to testing documented by industry observers, creative prompting can still bypass the filters to produce what have been described as “likeness-adjacent” characters, content that evokes a real person or copyrighted figure without technically reproducing them. It is a familiar challenge in AI governance: the gap between what a policy forbids and what a model can be coaxed into producing.

ByteDance’s vertical integration gives it a unique position in this contest. It builds the AI model, owns the editing platform where it is deployed, and controls TikTok, the dominant short-form video distribution channel. That control means it can, in theory, enforce IP protections across the entire pipeline from generation to distribution. Whether it will do so with sufficient rigour to satisfy Hollywood and its lawyers is another matter entirely.

The AI boom of 2025 produced a generation of tools that could generate text, images, and code at scale. Video was always the next frontier, and the hardest to govern. ByteDance’s bet is that it can be the company to commercialise AI video generation globally without drowning in litigation. The safeguards it has added to Seedance 2.0 are a necessary first step. Whether they are sufficient is a question that Hollywood, regulators, and policymakers across multiple jurisdictions will be answering for months to come.