← Back to stories Top view of scrabble tiles spelling 'DOCUMENTS' on various contracts and agreements.
Photo by RDNE Stock project on Pexels
虎嗅 2026-03-29

ByteDance's Seedance 2.0 overseas push hit by copyright storm; company reportedly tightening features and algorithms

Global rollout stalled

ByteDance (字节跳动)’s much‑lauded AI video model Seedance 2.0 has hit turbulence: its planned global release was paused after Hollywood rights holders, coordinated by the Motion Picture Association (MPA), warned of systemic copyright infringement. Seedance 2.0 had been hailed for film‑level motion logic, audio‑visual consistency and realism. How did a breakthrough technology get stopped in its tracks? Big studios say the model can reproduce iconic characters and scenes too faithfully; ByteDance has reportedly restricted some features and is adjusting internal algorithms while it addresses complaints.

How the fallout unfolded

The model’s February test rollout ignited industry excitement and investor moves, but user uploads quickly produced familiar and high‑fidelity recreations — from Marvel heroes to classic Chinese TV scenes — prompting sharp pushback. The MPA, representing Disney, Warner Bros. and other studios, sent formal letters accusing the system of training on copyrighted film and TV material without authorization and describing the infringements as systemic rather than accidental. It has been reported that Seedance 2.0 used a large amount of well‑known IP in its training data, and insiders say the model’s black‑box behavior made it hard for users to judge infringement risk.

Legal and industry reverberations

ByteDance has begun to impose domestic limits — users attempting to generate realistic human likenesses are met with copyright notices, and some reference‑by‑real‑person features have been suspended. This episode arrives amid an evolving legal landscape: Chinese courts have already found AI‑generated reuses infringing in a high‑profile 2025 case between Tencent and Baidu, and IP owners are increasingly demanding paid licensing or tighter controls. Experts warn that the core dispute is technical and legal: when does “learning” become unlawful copying, and who bears responsibility — model creator, platform or end user? It has been reported that some Chinese publishers are experimenting with licensing deals, offering a possible path toward “AI‑friendly” monetization.

What comes next

Seedance 2.0’s pause underscores a broader dilemma for AIGC — rapid technical progress colliding with unsettled copyright rules and heightened international scrutiny as Chinese AI firms seek global markets. Large firms may buy rights and offer “copyright shields” for enterprise customers, but that solution risks raising costs and concentrating power. Will clearer rules and licensing frameworks arrive before models scale overseas? The answer will shape whether AI video generation matures into a regulated commercial tool or remains a flashpoint between tech innovation and intellectual‑property enforcement.

AIPolicy
View original source →