Rooting for you

[kabachuha]

This model's video quality is very good. Imagine, what will happen if it gets audio generation support — it will simply obliterate the closed-source Wan2.5 et. at.

Even the Lite model will scare everything out of them.

And this is not so hard to do, for example — Ovi — from CharacterAI made it with not much resource usage by creating a twin backbone transformer, described in their paper https://arxiv.org/pdf/2510.01284, which resumed training from Wan2.2 5b.

Whether you are working on it already or are beginning to work, best of luck for you!

[MeiYi-dev]

Audio support and a 1-2 step distill like https://arxiv.org/abs/2510.08431 / https://arxiv.org/abs/2511.20549 / https://arxiv.org/abs/2510.27684 would be so nice!