wan2.2 inference steps vs self-force-plus inference steps

[DongaiLiu]

I found that in self_force_plus, the denoising model follows a process of 999 → 0, then adds noise back to 750 → 0. However, in wan2.2, it directly goes 900 → 750 → 500 → 250 → 0. Is there a difference between these two approaches?

[Doncic7]

我发现在self_force_plus中，去噪模型遵循999→0的过程，然后将噪声加回到750→0。但是在wan2.2中，它直接变为900→750→500→250→0。这两种方法有什么区别吗？

你好，可以分享下你基于self-forcing-plus修改的wan2.2蒸馏代码嘛