ByteDance Agrees Stronger Copyright Safeguards for AI Tools
ByteDance and Hollywood’s Motion Picture Association have set a framework for tighter copyright protections across Seedance, Seedream, TikTok, CapCut and Dreamina.
What you need to know
- ByteDance and the MPA have signed an MOU covering AI video and image-generation safeguards.
- The framework includes Seedance, Seedream and related services including TikTok, CapCut and Dreamina.
- The announcement confirms continued cooperation, but not a licensing deal, settlement or specific new penalties.
ByteDance has agreed a new framework with Hollywood’s Motion Picture Association (MPA) to strengthen copyright safeguards around its generative AI video and image tools. The companies announced the memorandum of understanding (MOU) on Monday 17 August, covering ByteDance models including Seedance and Seedream, along with services offered through TikTok, the TikTok USDS Joint Venture, CapCut and Dreamina.

The arrangement follows months of pressure from the film industry over AI-generated clips said to resemble protected characters, film material and actor likenesses. The MPA said the MOU establishes a shared framework for maintaining stronger protections for intellectual property owned by its member studios as ByteDance’s AI products develop.
It is important, though, not to read more into the announcement than has been disclosed. The MOU is a cooperation framework, rather than a published licensing deal or court settlement. Neither side has confirmed a financial agreement, penalties for future infringement, changes to training data, removal of material from already-trained models, or a list of blocked characters and franchises.
A formal response to earlier concerns
The dispute began after Seedance 2.0 launched in China on 12 February. The MPA said the video-generation service could produce material featuring copyrighted characters and other protected works without meaningful safeguards. One widely circulated example reportedly depicted Tom Cruise and Brad Pitt fighting in an AI-generated video.
On 20 February, the MPA sent ByteDance a cease-and-desist letter concerning Seedream 5.0 Lite and Seedance 2.0. The association accused ByteDance of making content that closely replicated distinctive characters and other protected elements from member studios’ works, and demanded safeguards intended to prevent infringing outputs.
ByteDance subsequently said it was taking steps to reinforce its protections. Its general counsel, John Rogovin, said: “ByteDance respects the intellectual property rights that underpin creative industries around the world, and we believe responsible innovation in AI goes hand in hand with meaningful protections for rightsholders.”
MPA chairman and chief executive Charles Rivkin said the organisations had spent several months in “constructive engagement” over meaningful guardrails for Seedance and Seedream. “Today’s agreement illustrates our belief that copyright is a cornerstone of the film and television industry – and reinforces our commitment to protect creative content,” Rivkin said.
What protections are already disclosed?
ByteDance’s Dreamina materials describe a set of existing safeguards across its generative tools. These include moderation of prompts, reference inputs and generated material; controls designed to prevent unauthorised use of likenesses and intellectual property; proactive monitoring; in-app reporting; and direct collaboration with rightsholders.
The company also says its tools can apply visible AI labels, invisible watermarking and C2PA Content Credentials. Those credentials are intended to help platforms identify material made using AI. TikTok has separately said real-face uploads are blocked in the relevant Symphony workflow, with safeguards applying to image, video and audio references.
The MPA said ByteDance’s newer Seedream 5.0 Pro and Seedance 2.5 releases show continued progress on intellectual-property protections. It did not, however, publish technical evidence or test results showing how much infringement has been reduced, nor did it detail the precise measures introduced following its feedback.
Seedance and Seedream are becoming more capable
The timing matters because ByteDance is expanding what its creative AI models can do. Seedance 2.5 is designed for up to 30-second video storytelling, according to ByteDance’s Seed model listings, and includes reference control and editing features. Dreamina says that is double the previous 15-second maximum, while the model can accept up to 50 multimodal references including images, video, audio and text.
Dreamina also says Seedance 2.5 offers more granular editing and timestamp-specific controls. Seedream 5.0 Pro, meanwhile, is described as a multimodal image-generation model with advanced reasoning and content-creation capabilities.
Seedance 2.5 has begun rolling out to subscriber accounts for people over 16 in Europe, Asia, the Middle East and South America, according to Dreamina. A specific UK launch date and separate UK eligibility statement have not been confirmed. Dreamina says Seedream 5.0 Pro is available to paying subscribers in Europe and several other regions, but a UK price has not been confirmed.
What happens next?
The key development is that the MPA and ByteDance now have a formal route for continued cooperation across products and platforms. The MPA says it continues to monitor for copyright infringement, while ByteDance says it will keep strengthening safeguards across its generative AI services.
For users, the practical effect may be more moderation around prompts, reference images and generated results, particularly where requests closely imitate studio-owned characters, recognisable film elements or real people. But the boundaries of those restrictions remain unclear. The MOU does not spell out which requests will be blocked, how disputes will be handled, or what data ByteDance may share with the MPA.
That leaves the agreement as a significant policy signal rather than a fully transparent technical roadmap. As video AI becomes more capable and accessible through consumer-facing services, the test will be whether the promised guardrails are consistent, visible to users and effective when creators try to push beyond them.
Why it matters
For UK users, stronger guardrails could make it harder to generate clips or images resembling protected film characters, studio material or real people without permission. The agreement also underlines a growing expectation that AI tools should identify generated material and give rightsholders routes to report misuse. What has not been confirmed is exactly how ByteDance will change its models, training data or output restrictions.
