Studios challenge ByteDance over AI training

A coalition of major Hollywood studios and industry bodies has accused ByteDance of exploiting copyrighted film and television material to train its artificial intelligence video tools without permission, intensifying a widening legal confrontation between the entertainment sector and generative AI developers.

The dispute centres on ByteDance’s fast-evolving video generation systems, integrated into platforms linked to TikTok and its editing suite CapCut, which allow users to create cinematic clips from text prompts. Studio executives and guild representatives contend that the underlying models were trained on protected screen content, raising concerns that scenes, visual styles and character likenesses could be replicated without licensing agreements or compensation.

Representatives of the coalition argue that the unauthorised use of copyrighted works undermines the economic foundations of the film and television industry. They maintain that AI systems capable of synthesising scenes in the style of well-known franchises or directors threaten established production models, particularly if generated clips can be monetised on social media platforms at scale.

ByteDance has not publicly disclosed the full scope of the datasets used to train its video models. The company has previously stated that it respects intellectual property rights and complies with applicable laws, while emphasising that generative AI systems are trained on a mixture of licensed data, publicly available material and user-generated content. It has also highlighted internal safeguards designed to prevent users from creating infringing or harmful outputs.

The clash reflects broader tensions that have been building across the creative industries since the rapid commercialisation of generative AI in 2023. Technology firms including OpenAI, Stability AI and Midjourney have faced lawsuits from authors, visual artists and media companies alleging the unauthorised scraping of copyrighted material to train large models. Courts in the United States and Europe are now grappling with complex questions around fair use, transformative works and the boundaries of machine learning.

Within Hollywood, anxiety over AI training practices formed a central plank of last year’s labour negotiations. The Writers Guild of America and SAG-AFTRA secured contractual protections limiting the use of AI to generate scripts or digital replicas of performers without consent. Industry leaders view the ByteDance dispute as part of the same continuum, arguing that unchecked training practices could dilute creative control and erode residual income streams.

Legal analysts note that any formal case against ByteDance would hinge on demonstrating that identifiable copyrighted works were copied during the training process and that resulting outputs are substantially similar to protected content. That task is complicated by the opaque nature of large-scale data ingestion and the probabilistic way in which generative models assemble new images from patterns learned across vast datasets.

Technology companies, for their part, contend that training AI models on large corpora constitutes a transformative process analogous to search indexing, and therefore may qualify as fair use under US law. They argue that prohibiting such training would stifle innovation and entrench dominant players with access to proprietary data.

The entertainment coalition counters that generative video tools differ fundamentally from search engines because they can reproduce expressive elements of original works, potentially substituting for licensed productions. Executives point to examples in which AI-generated clips appear to echo the visual tone, character design or narrative arcs of established films and television series.

Policy makers in Washington and Brussels are monitoring the unfolding disputes as they consider new regulatory frameworks for artificial intelligence. Proposed legislation in the European Union already includes transparency requirements around training data, while lawmakers in the United States have floated measures aimed at clarifying copyright liability in AI development.

Market dynamics add another layer of complexity. ByteDance commands a vast global user base through TikTok, giving its AI video tools immediate distribution channels that traditional studios lack. Analysts say that if generative video becomes a mainstream content format, social media platforms could emerge as direct competitors to conventional film and television producers.

At the same time, some studio executives acknowledge that AI technology offers efficiencies in pre-visualisation, special effects and localisation. Several major production houses have invested in in-house AI capabilities or partnered with technology firms to streamline workflows. The current dispute, they stress, is less about rejecting innovation than about establishing clear rules governing data use and compensation.

Intellectual property scholars suggest that the outcome of any litigation could reshape the balance of power between technology platforms and content owners. A ruling favouring studios might compel AI developers to negotiate licensing deals, creating a new revenue stream for rights holders. A decision siding with AI companies could entrench broad fair use interpretations, accelerating the proliferation of generative tools.