对比：Seedance 2.5 仍压过 Kling 4.0 与 Wan 3.0

Luke Tsiaras @LukeTsiaras原推发布于 2026.10.22

在 X 上打开原推

Higgsfield has been lying to us. Wan 3.0 is nowhere close to Seedance 2.5. After watching way too many video generation comparisons, it’s become increasingly obvious to me just how impressive Seedance 2.5 really is. Even today, nothing comes close. Not Wan 3.0, not even the newly released Kling 4.0. Neither can compete with the quality, versatility, and consistency of Seedance 2.5. Seedance has always had an edge over the rest of the market, but Seedance 2.5 may have been one of the greatest step-function improvements, and one of the closest things to an AGI moment, we’ve seen in video generation. The fact that a model this old is still so dominant is genuinely impressive. I have absolutely no idea what to expect from Seedance 3.0. But if I had to make one prediction, it’s that Hollywood actors should probably start protesting now before that model even releases. At the rate these models are improving, and given how easily AI-generated content can already capture and retain human attention, I think demand for traditional actors is going to diminish rapidly. (I’ll have a more thorough article on this coming soon.) @alexmashrabov @codentropy @Mahi