What did the MPA and ByteDance actually agree to?

They signed a memorandum of understanding setting a restriction framework for how ByteDance's generative AI video and image models may draw on studio film and television properties. It covers TikTok, the TikTok USDS joint venture, CapCut and Dreamina, and names the Seedance 2.0 and 2.5 and Seedream 5.0 Lite and 5.0 Pro models.

Can I still generate videos of movie characters on TikTok or CapCut?

Expect it to get harder. The framework governs how these models pull from existing film and TV properties, and enhanced protections have already shipped in Seedance 2.5 and Seedream 5.0 Pro. The exact filtering method has not been published, so the practical limits will only be clear as people hit them.

Why did the Motion Picture Association target ByteDance?

In February 2026 it sent ByteDance a cease-and-desist letter over Seedream 5.0 Lite and Seedance 2.0 after users prompted the tools to regenerate copyrighted studio material. One widely circulated example generated a fight scene between Tom Cruise and Brad Pitt.

Will my existing AI videos be deleted?

Nothing in the announcement says so. The memorandum concerns what the generative models produce going forward, not a removal sweep of content already uploaded. Normal copyright takedown processes on these platforms are unaffected and continue to apply separately.

Is this agreement legally binding on ByteDance?

A memorandum of understanding is a negotiated framework rather than a court order, and the announcement did not detail enforcement mechanisms or an implementation timeline. It resolved the dispute without litigation, but the published terms do not describe penalties for failing to meet it.

TikTok and CapCut Get New AI Copyright Guardrails

Which apps and models does the deal cover?

The agreement applies to ByteDance's two generative model families by name: Seedance 2.0 and 2.5 for video, and Seedream 5.0 Lite and 5.0 Pro for images. Those models sit behind the AI generation features in TikTok, CapCut and Dreamina, so the restriction framework reaches the consumer apps rather than just a back-end research model.

Both sides describe it as a shared framework for keeping guardrails on those models, with the stated goal of protecting intellectual property. The association's chief executive, Charles Rivkin, framed the deal around copyright as "a cornerstone" of the industry. ByteDance general counsel John Rogovin said the company respects the intellectual property rights that underpin the creative industries.

Why did Hollywood go after ByteDance?

This did not start as a negotiation. In February the association sent ByteDance a cease-and-desist letter over Seedream 5.0 Lite and Seedance 2.0, and the complaint was specific: users were prompting the tools to regenerate recognisable studio material. One widely shared clip generated a fight scene between Tom Cruise and Brad Pitt.

That is the pattern the studios have been fighting across every AI video tool — not the technology itself, but a text box that will happily rebuild a protected character or a famous performer's likeness on demand. Six months of pressure later, the two sides have a written framework instead of a lawsuit.

According to the announcement, enhanced protections have already been rolled out in the newer Seedream 5.0 Pro and Seedance 2.5 releases, which suggests some of the filtering is live rather than promised.

What changes if you make AI videos on TikTok?

The honest answer is that the announcement does not spell this out, and that gap matters. What is documented:

Studio characters and footage get harder to generate. The framework governs how the models may pull from existing film and TV properties, so prompts aimed at protected characters are the obvious target.
It applies inside the apps you already use. CapCut and Dreamina are editing tools with large casual user bases, not niche AI services.
Nothing announced changes existing uploads. The MOU concerns what the models generate, not a sweep of what is already posted.

What has not been disclosed is more interesting: the exact enforcement mechanism, whether individual studios can add their own titles to a blocklist, the implementation timeline, and any user-facing policy change. None of that was detailed, and it would be wrong to assume a specific technical method has been agreed.

Is this a bigger shift for streaming platforms?

It fits a pattern of studios choosing to license and channel fan-made AI content rather than only block it. Disney has already been running a US pilot that pulls TikTok fan videos into Disney+, and generative tools are moving into mainstream production pipelines — NVIDIA now markets DLSS 5 as bringing Hollywood-grade rendering to games.

The commercial logic is straightforward. Studios want their catalogue protected on platforms where hundreds of millions of people watch short video, while still wanting to be present there. An agreement that restricts generation but keeps the relationship intact serves both goals better than a court fight, which is why this landed as a memorandum rather than a filing.

If you use AI generation inside these apps, the practical takeaway is to expect more refusals on prompts naming films, shows and performers, and to stop assuming a tool's willingness to produce something implies you are clear to post it. For alternatives that keep their rules in plain sight, our roundup of the best AI apps for iPhone is a reasonable place to start, and the platform side of streaming keeps shifting quickly — Netflix has just dated The Diplomat season 4.