CourionAI
EN
Newsletter
← All news
copyright 2 min read

Hollywood and TikTok's Owner Signed a Truce on AI Video, but Not the Part That Matters Most

The Motion Picture Association and ByteDance agreed a global framework for IP protection in Seedance and Seedream. It covers filters and content credentials, not whether training on copyrighted film was legal.

A strip of film passing through a filtered gate, one frame carrying a wax seal, a wide door still standing open behind

The Motion Picture Association, the trade body for the big US studios and streamers, signed a memorandum of understanding with ByteDance on Tuesday setting out a global framework for protecting intellectual property in AI video and image generation. It covers ByteDance’s Seedance video model and Seedream image model wherever they appear, which now means TikTok, the TikTok USDS joint venture, CapCut and Dreamina.

The backstory explains the urgency. When Seedance 2.0 launched, users quickly produced clips featuring recognisable riffs on actors including Brad Pitt and Tom Cruise. The MPA sent ByteDance a cease-and-desist in February over Seedream 5.0 Lite and Seedance 2.0, and months of talks followed. MPA chairman Charles Rivkin framed copyright as a cornerstone of the industry. No licensing fees were disclosed, which tells you what kind of agreement this is: a truce, not a payment deal.

Here is the distinction worth holding on to, because it is the whole story. There are two separate copyright questions in generative AI. The first is the output question: does the model let you produce something that looks like a specific film, character or actor? That is fixable with engineering, and it is what this agreement addresses through content filters, face blocking, and C2PA Content Credentials, an industry standard that attaches tamper-evident metadata to a file recording how it was made. The second is the input question: was it lawful to train on copyrighted films and images in the first place? This MOU explicitly does not settle that. Those disputes stay where they were, in court and in legislatures.

That is not a criticism so much as an observation about how these deals tend to work. Output controls are the part both sides can agree on quickly, because a studio wants its stars protected and a platform does not want lawsuits over what its users generate. The input question is worth vastly more money to whoever wins it, so nobody trades it away in a memorandum. Also worth noting: an MOU is a statement of intent, not a contract with penalties attached.

What this means for you: if you make videos, this is the direction of travel. Expect the big generators to get progressively stricter about producing recognisable actors, characters and franchises, and expect more of your output to carry Content Credentials whether or not you asked for them. That is mostly good news, because provenance metadata also helps you prove what you made yourself. If you have been using AI video tools to make things that lean on existing IP, treat that as a shrinking window rather than a stable practice. And if you are just watching from the sidelines: the interesting fight, the one about training data, has not been settled by this at all.

Sources

Source: https://www.hollywoodreporter.com/business/digital/mpa-inks-ai-video-ip-protection-bytedance-1236675016/

Next story

OpenAI Froze Its Own Training for Two Weeks, and Now Watches Its Models Like a Hawk

After the July Hugging Face incident, OpenAI paused reinforcement learning for two weeks and built a monitoring system meant to raise an alarm within 30 minutes. It costs about 20 percent extra compute.

A walled maze with one sealed gate, a large watchful lens hovering above it and a small alarm bell at the edge