Seedance 2.0 might be gen AI video’s next big hope, but it’s still slop

The Verge
Seedance 2.0 shows impressive video generation but remains 'slop' due to its reliance on potentially infringing training data.

Summary

Seedance 2.0, ByteDance's new video generation model, has produced highly impressive clips, such as one featuring a digital Tom Cruise, alarming Hollywood studios who sent cease and desist letters over copyright infringement claims. While the footage looks superior to competitors like Sora and Veo, the article argues that its primary fame comes from generating polished ripoffs, labeling it a 'fancy slop generator.' The core issue is that the model's ability to mimic reality stems from training on vast, potentially unauthorized, visual data, making it fundamentally similar to its peers despite better aesthetics. A short film by Jia Zhangke, generated with Seedance 2.0, demonstrated narrative cohesion but still exhibited continuity errors common to all video generators, suggesting that skilled users can work around limitations. Ultimately, the article concludes that for AI video to shed the 'slop' association, companies must prove their models can create quality content without relying on stolen intellectual property, a challenge companies like Asteria and Adobe are addressing with IP-safe models.

(Source:The Verge)