China takes the video-model lead

Artificial Analysis runs a video arena the way a taste test runs: show two clips from the same prompt, hide which model made each, count which one people pick. By late May, the top of its text-to-video ranking was almost entirely Chinese — ByteDance's Seedance at first, Kuaishou's Kling just behind, and seven of the top eight models from Chinese labs. OpenAI's Sora and Google's Veo, the names that taught the public what AI video was, landed around eighth and ninth.

The copyright violations are “baked into the technology itself.” — the Motion Picture Association, on Seedance 2.0

The usual explanation is data. ByteDance trained Seedance on Douyin, the Chinese sibling of TikTok; Kling learned from Kuaishou, a separate short-video platform with its own ocean of clips. Years of people filming each other gave these labs a training set Western firms can't easily assemble — though cost matters too: the Chinese tools undercut US pricing many times over and ship inside editing apps creators already use.

The same data that won the leaderboard is now the legal problem. Seedance reproduced Marvel and Star Wars characters closely enough that, after cease-and-desist letters from Disney, Paramount, Warner Bros., Netflix and the film industry's trade body, ByteDance suspended its global launch in March. The film lobby's argument is that the infringement is baked into the model itself, not into how people use it. So the clip the world voted best is, outside China, barely available to try.