Trained to Imagine
Rendering
Creative Intelligence
AI Production / Hollywood / Legal

The MPA and ByteDance agree IP protections for Seedance and Seedream

Source: Variety ·

The Motion Picture Association and ByteDance signed a memorandum of understanding on IP guardrails for the Seedance video and Seedream image models across TikTok, CapCut, and Dreamina. It follows the MPA's February cease-and-desist over Seedance 2.0; the specific safeguards and enforcement terms were not disclosed.

TTI Takeaway

The studios are moving from suing model makers to negotiating with them — at least the ones with consumer platforms to protect. Undisclosed terms mean nobody outside can check them, so for client work the practical rule does not change: guardrails in the model are not clearance for the output.

Analysis

Six months ago the MPA was accusing ByteDance of disregarding copyright law “on a massive scale.” Now the two have signed a memorandum of understanding and issued matching statements about constructive engagement. The speed of that turn is the story. It shows what the studios can extract from a model maker that has consumer platforms to protect, and it suggests a template that may be easier to reach than a court ruling.

According to Variety, the cease-and-desist in February followed the release of Seedance 2.0, when users produced riffs on actors including Brad Pitt and Tom Cruise. The two sides then began talks, and the protections they developed were, in their words, reflected in Seedance 2.5 and Seedream 5.0 Pro, released the month before the announcement. The framework covers ByteDance’s generative products across TikTok, its US variant, CapCut and Dreamina. MPA chairman Charles Rivkin described the aim as continuing to “further fortify those guardrails,” and ByteDance general counsel John Rogovin called the MOU “an important framework for continued collaboration.” Variety notes the MPA took a similar line with OpenAI’s Sora before that company added its own guardrails.

What is missing is everything a third party would need to evaluate it. The specific safeguards, how they are tested, what happens when they fail, and whether there is any enforcement mechanism were not disclosed. An MOU is also, by nature, a statement of intent rather than a licence. It does not tell a user what they are permitted to make, and it does not transfer any rights to the output.

For productions and agencies, the practical consequence is that nothing about clearance changes. A model that refuses to render a famous actor’s face or a franchise character is reducing obvious risk at the prompt; it is not certifying that an output is free of protected elements. Look-alikes, costumes, distinctive sets and score-adjacent audio can all get through a filter. The review still has to happen on the output, by people who know what they are looking for, and the vendor’s terms still determine who carries the risk. This is the same gap the Midjourney case exposed in June, when the defence sought discovery into the studios’ own AI use: the rules for what counts as acceptable use are being set in private negotiations and litigation, not in public standards.

The next thing to watch is whether this becomes a pattern for other providers with large consumer platforms, and whether any of these agreements ever publish their terms. Studio negotiators have every reason to keep details private, and model makers have every reason to cite the deal as reassurance. Our view is that guardrails negotiated out of sight are a useful reduction in risk for the studios that negotiated them, and only that. For everyone else, the output still has to be cleared.