Generative artificial intelligence is reaching a new milestone in audiovisual creation. Following OpenAI’s Sora and Google’s Veo, it is now the turn of ByteDance, TikTok’s parent company, to shake up the industry with Seedance 2.0, a video model launched in China on February 12, 2026. Capable of producing scenes of spectacular realism from a simple text prompt, the tool is sparking both fascination and concern. In Hollywood, some are already calling it a historic turning point for the film industry.
According to several online demonstrations, Seedance 2.0 is said to be capable of generating sequences featuring well-known actors or franchises, with a level of visual detail and narrative coherence rarely achieved before. Unlike its Western competitors, which often have strict filters on faces or intellectual property, the Chinese platform appears to take a more permissive approach. This strategic difference is fueling trans-Pacific tensions.
A new generation of ultra-realistic video AI
AI-powered video creation is experiencing rapid growth. According to McKinsey, more than 30% of companies in the media and entertainment sectors are already experimenting with generative AI tools in theirproduction workflows¹. The global generative AI market could exceed $100 billion by2030², with video being the fastest-growing segment.
Seedance 2.0 is part of this trend. Based on textual instructions, the model generates animated sequences featuring realistic settings, fluid camera movements, and consistent lighting. The demos published online showcase scenes reminiscent of superhero universes, medieval sagas, or Hollywood action movies.
This realism can be attributed to recent advances in video diffusion and multimodal models, which are capable of learning simultaneously from text, images, andtemporal sequences³. Advances in big data processing and computing power now make it possible to train these models on vast amounts of audiovisual content.
Viral videos that are causing a stir in Hollywood
Content created through Seedance 2.0 is going viral on social media. Users have shared fictional battles between heroes from different universes or imagined alternative scenes for cult TV shows.
The viral nature of this content raises a key question: Who owns the rights to these images?
The U.S. audiovisual industry is worth over $700 billionglobally⁴. Disney alone generated nearly $89 billion in revenue in2023⁵. The possibility that AI could be used to recreate, repurpose, or extend copyrighted works without authorization poses a significant economic risk.
Actors’ and screenwriters’ unions, which were already active during the 2023 strikes over the use of AI, fear that unauthorized reproductions will become commonplace. Academic studies highlight that deepfakes and video-generation systems pose major legal challenges regarding image rights andintellectual property⁶.
ByteDance promises safeguards
In response to criticism, ByteDance asserts that it respects copyright and announces plans to strengthen its protection mechanisms. However, the company has not provided details on the nature of the training data used for Seedance 2.0, nor on the specific filtering methods implemented. This lack of transparency fuels mistrust. Recent research on the governance of generative models shows that transparency regarding datasets is a key factor in preventingillicit use⁷.
At the same time, Seedance 2.0 offers a tiered pricing structure, ranging from $41 per month for the Basic plan, to $83 for the Pro plan, up to $167 per month for the Max plan, with each plan offering an increasing number of video generation credits. This structure clearly positions the tool as a solution designed for regular, even intensive, production, rather than as a mere experimental tool.
At this stage, Seedance 2.0 appears to highlight a strategic divergence between Western models—which are more constrained by regulation and industry pressures—and Chinese models operating within a distinct regulatory framework, where rapid deployment and economic competitiveness are seen as top priorities.
A technological and industrial breakthrough
Beyond the media buzz, Seedance 2.0 represents a more profound shift: the democratization of audiovisual production.
Producing a scene with cinematic quality has traditionally required large crews, extensive infrastructure, and massive budgets. If AI can achieve comparable results at a lower cost, it could transform the industry’s economic landscape.
Researchers in the field of innovation economics estimate that creative automation could reduce certain production costs in the cultural industries by up to 20% inthe mediumterm⁸. However, this increased efficiency comes with a risk of disintermediation for creators.
Ethical and Legal Issues
Seedance 2.0 raises several major questions:
- Potential infringement of copyright and the right of publicity
- The risk of hyper-realistic deepfakes
- Difficulties in determining liability and legal responsibility
- Pressure on employment in the creative industries
Regulation of AI is advancing, particularly in Europe with the AIAct⁹, but cross-border enforcement remains complex. Against the backdrop of global technological competition, the balance between innovation, sovereignty, and the protection of creators remains fragile.
A new chapter in the global AI war
Seedance 2.0 is not merely a technical advancement. It symbolizes a geopolitical battle over control of advanced generative models. Following the race to develop large language models, the competition is shifting toward video and immersive environments.
Hollywood is watching this rise with concern. Generative AI is entering a phase where the line between human creation and algorithmic synthesis is becoming almost indistinguishable.
It remains to be seen whether the industry will opt for legal action, strategic integration, or a combination of the two.
How does Seedance 2.0 work?
Seedance 2.0 is built on a text-to-video architecture based on spatio-temporal diffusion models. Unlike traditional image generators, the model does not merely predict pixels; it models a dynamic probabilistic distribution over time to ensure consistency between successive frames.
- A multimodal encoder that transforms text prompts into conditional latent representations
- A video diffusion model operating in a compressed latent space (latent diffusion), optimized for temporal continuity
- Spatio-temporal attention mechanisms that maintain facial identity, scene stability, and movement coherence
- A high-resolution decoding phase aided by super-resolution networks
Learn more
This rise of generative models in imaging and video is part of an intensifying competition between tech companies and the creative industries. On a related topic, check out our article “Meta x Midjourney: A Strategic Alliance to Revolutionize AI-Powered Images and Video”, which analyzes how technology partnerships are reshaping the balance between algorithmic innovation and audiovisual production.
References
1. McKinsey & Company. (2023). The economic potential of generative AI.
https://www.mckinsey.com
2. Bloomberg Intelligence. (2024). Generative AI Market Outlook 2030.
https://www.bloomberg.com
3. Ho, J. et al. (2022). Video Diffusion Models. arXiv.
https://arxiv.org/abs/2204.03458
4. Motion Picture Association. (2023). Theme Report 2023.
https://www.motionpictures.org
5. The Walt Disney Company. (2023). Annual Report.
https://thewaltdisneycompany.com
6. Chesney, R., & Citron, D. (2019). Deepfakes and the New Disinformation War. Foreign Affairs.
https://www.foreignaffairs.com
7. Bommasani, R. et al. (2021). On the Opportunities and Risks of Foundation Models. Stanford CRFM.
https://crfm.stanford.edu
8. Bakhshi, H., & Higgins, D. (2022). Automation and the creative industries. Nesta.
https://www.nesta.org.uk
9. European Parliament. (2024). Artificial Intelligence Act.
https://www.europarl.europa.eu