Product2 distinct publishers3 min readUpdated
The memorandum covers every ByteDance AI model, including those in CapCut and Dreamina. The terms are undisclosed, so builders get a rulebook they cannot inspect.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
ByteDance and the Motion Picture Association have signed a memorandum of understanding that the MPA, in a press release, calls a shared framework for protecting its members against copyright infringement [1]. Neither side has spelled out what the safeguards actually are [2], which means the people shipping on CapCut and Dreamina now work under a rulebook they cannot read.
The scope is broad. According to the MPA, the memorandum applies to all ByteDance AI models, including those used on TikTok, the TikTok USDS Joint Venture, CapCut and Dreamina [3], four named product surfaces [4]. The models at the centre of the dispute are Seedance, ByteDance's video generator, and Seedream, its image counterpart, which sit behind the AI features in TikTok, CapCut and Dreamina [5]. The MPA's members include Disney, Paramount and Warner Bros. Discovery [6]. The trade body said the agreement "represents significant cooperation between the two organizations" [7] and that Seedream 5.0 Pro and Seedance 2.5 reflect "continued advances in IP protections" [8].
That is a warm read of a fight that opened in February, when the MPA sent a cease-and-desist letter after Seedance 2.0 appeared, calling the tool an engine of "systemic infringement" [9]. Individual studios including Disney, Netflix, Warner Bros, Paramount and Sony had already sent their own warnings [10]; Disney pointed to viral clips featuring Marvel and Star Wars characters, and Japan opened a separate inquiry into anime look-alikes [11]. ByteDance then pledged to strengthen its safeguards around unauthorised use of intellectual property and likeness, and reportedly suspended Seedance 2.0's global rollout [12]. MPA chief executive and chair Charles Rivkin has since described "constructive engagement" and said the company implemented "meaningful guardrails" [13]. The version numbers are the tell: the model that drew the letter was 2.0, the model now cited as evidence of progress is 2.5, half a release later [14].
The MPA's complaint had two prongs, one about outputs, that Seedance would generate recognisable copyrighted characters and celebrity likenesses on demand, and one about inputs, that the model was trained on studio content without permission [15]. The pact answers the first far more than the second [16]. Output filters are the tractable half: making a model decline to draw Iron Man is an engineering task, and it settles nothing about whether the film libraries taught it what Iron Man looks like [17]. The studios also are not obviously being paid, because a voluntary pact is not a licensing deal [18].
For operators, the practical problem is asymmetry of information. The specifics are undisclosed, which makes the guardrails impossible to assess from outside and easy to loosen quietly, and enforcement across TikTok, CapCut and Dreamina at those apps' scale is a formidable undertaking [19]. A prompt or template that clears review this month may fail next month, and there is no published standard to appeal to.
Context for how much leverage this represents: Hollywood has watched OpenAI pull its Sora app amid legal and reputational heat and seen the Oscars ban AI actors and AI-written scripts [20], while studios remain in court against Midjourney on infringement claims [21]. Litigating ByteDance, headquartered beyond easy reach of US courts, was an uncertain prospect, so the MPA negotiated guardrails it can point to now [22].
Watch whether any of the terms are published, whether the MPA carries this template to rival generators [23], and whether the truce hardens into the licensing regime the studios actually want [24].
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
The MPA announced in a press release that it signed a memorandum of understanding with TikTok parent ByteDance detailing a shared framework to protect against copyright infringement of the trade group's members.
ByteDance says newer versions of the models already carry stronger intellectual-property safeguards, but neither side has spelled out what those guardrails actually are.
The MPA said "this agreement represents significant cooperation between the two organizations."
In February, after Seedance 2.0 appeared, the MPA sent a cease-and-desist letter calling the tool an engine of "systemic infringement" and accusing ByteDance's AI models of using copyrighted material without permission.
A voluntary pact is not a licensing deal, so the studios are not obviously being paid for whatever their catalogues may already have taught the models.
The specifics are undisclosed, making the guardrails impossible to assess from the outside and easy to loosen quietly, and enforcement across TikTok, CapCut and Dreamina at the scale those apps operate is a formidable undertaking.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Agreement confirmed, substance unpublished
Two independent publishers confirm a signed memorandum and its stated scope, and one primary artefact (the MPA press release) underpins the announcement. But the operative content - what the guardrails are, how they are tested, what happens on breach - is undisclosed by both parties, and the training-data allegation is unresolved, so the evidence supports the existence of a deal far better than its effect.
Broad nominal scope, unverified implementation
Nominal reach is wide: the MPA says the memorandum applies to all ByteDance AI models across four named surfaces, and ByteDance has already shown willingness to act on model availability by reportedly suspending Seedance 2.0's global rollout. Against that, no source provides any measurement of guardrail deployment, filtering behaviour, or enforcement coverage across TikTok, CapCut and Dreamina, so uptake beyond the signature and version releases is unobserved.
Announcement language outruns verifiable substance
The MPA's public framing - "significant cooperation," "meaningful guardrails," "continued advances in IP protections" - and the framing of ByteDance as a template for rival generators sit above what the record shows: undisclosed terms, an output-side remedy, no licensing or disclosed compensation, and an unresolved training-data claim while studios still litigate Midjourney. The overstatement comes from the parties' own characterizations rather than from fabricated facts, and TNW's coverage partially discounts it.
Both signatories benefit from the announcement
The public record here originates with an MPA press release, and both parties gain from it: the MPA converts a stalled enforcement fight into a citable win and a precedent to press on other generators, while ByteDance defuses studio and regulatory pressure - cease-and-desist, individual studio warnings, a Japanese inquiry - without conceding licensing or training-data liability. The absence of disclosed terms means neither side bears verification risk for its claims.
Facts consistent across two outlets, substance unverifiable
Two publishers reporting the same day agree on the parties, timing, covered models and the February cease-and-desist history, which supports confidence in the factual spine. Confidence is capped because the cluster contains only these two accounts, one key detail (the Seedance 2.0 suspension) is reported second-hand, ByteDance is not directly quoted, and the memorandum text is unavailable to either outlet.
product
The Williams estate's answer to AI slop is a live Instagram account, not a takedown queue2 distinct publishers
product
Rogue Studio ships an uncensored video model and prices permission from $29 to $2991 distinct publisher
leadership
AI sponsorships now carry a 20% to 30% surcharge for the comment section1 distinct publisher
invest
Meta's Oakland trial is the only one of three megacap legal fronts with a number attached1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 17, 2026
1 article · August 17, 2026