In just a few years, artificial intelligence has transformed the way images are produced, distributed, and interpreted. We are witnessing a historical transition in which the ability to generate images no longer necessarily depends on a film crew, a camera, or a physical set. A single line of text can now be enough.
This transformation, however, is not only technological. It concerns the very infrastructure of the contemporary imagination: who owns images, who can reuse them, and, most importantly, who controls their future.
The recent case involving ByteDance and the video-generation model Seedance 2.0 has become one of the clearest points of friction between two cultural systems that, until a few years ago, seemed largely separate: the traditional film industry and the emerging ecosystem of generative artificial intelligence.
When the Dataset Meets Hollywood
On February 12, 2026, ByteDance introduced Seedance 2.0 in China, a multimodal system designed to transform text prompts, images, audio, or video into complete cinematic sequences. The ambition of the project was clear: to bring generative audiovisual production to an industrial scale, targeting not only experimental users but also professional sectors such as advertising, e-commerce, and film production.
The technology operated according to a logic that has become familiar within the AI landscape: a system trained on vast visual archives capable of synthesizing new images based on statistical patterns learned during training.
The problem, as often happens in these cases, was not technical but cultural.
Only a few hours after the launch, a short video generated with Seedance began circulating online. The clip featured Brad Pitt and Tom Cruise engaged in a rooftop fight scene, according to Irish director Ruairi Robinson, the sequence had been created using only two lines of prompt text.
Within hours the video accumulated millions of views. Yet instead of being interpreted as a technical demonstration, it was perceived by the film industry as a warning signal.
The Suspicion: An Invisible Library of Protected Images
The first official reaction came from Disney.
In a formal letter sent to ByteDance the day after the launch, the studio accused the company of building Seedance using a massive unauthorized library of copyrighted material. According to Disney, the system appeared capable of generating characters from franchises such as Star Wars and Marvel with a level of precision that suggested the direct use of proprietary material within the training dataset.
The issue did not concern only the potential misuse by end users. The point raised by the studio was more fundamental.
If a model has been trained on archives containing protected intellectual property, then the infringement does not occur at the moment an image is generated, but within the very structure of the system itself.
In other words, the infrastructure of artificial intelligence could already incorporate illegitimate content even before the first image is produced.
The Film Industry Closes Ranks
In the days that followed, other major studios began to respond.
Warner Bros. Discovery sent a formal letter to ByteDance accusing the company of infringing characters belonging to the DC universe, as well as content derived from series such as Game of Thrones. The document argued that the issue could not be attributed to users writing prompts, but rather to the design of the model itself.
According to the studios, Seedance appeared to reach the market already predisposed to produce content traceable to specific intellectual properties.
Within a few days Paramount, Sony, and Netflix adopted the same position, while the Motion Picture Association (MPA) sent its first official letter to a major generative AI company.
The message was clear:
this was no longer about isolated viral examples, but about a potential systemic precedent.
The association stated that the repetition and consistency with which protected content appeared in the outputs indicated something more than a simple technical error. According to the MPA, the infringement was not an accidental defect but a predictable consequence of the model’s structure.
The Suspension of the Global Launch
Facing growing legal pressure, ByteDance decided to suspend the international launch that had been scheduled for March 2026. The model remains available in China through the Volcano Engine platform, but the global rollout has been frozen while the company’s legal team and engineers work on new protection mechanisms. The company stated that it intends to strengthen safeguards preventing the generation of protected characters and to improve controls over the use of human identities and likenesses.
This decision follows an earlier intervention: the suspension of the voice-cloning feature, which had allowed realistic voices to be recreated from a single image of a person’s face. The Seedance case therefore illustrates how the challenges surrounding AI extend beyond images to include identity, voice, and digital presence.
A particularly revealing comparison emerges when examining the strategy adopted by other companies in the sector. When OpenAI introduced Sora, its video-generation model, tensions with Hollywood were addressed through negotiations and licensing agreements. This process eventually led to a collaboration with Disney that authorized the use of more than two hundred characters from the studio’s catalog.
This represents a fundamentally different model from the situation surrounding ByteDance. In the first case, artificial intelligence becomes a new infrastructure for the distribution of licensed content. In the second, it appears more like a machine recombining images drawn from undeclared archives. The difference is not technological but economic: it concerns the control of rights.
