On: Stable Cascade

[GaggiX]

I think that this model used consistency loss during training so that it can yield better results with less steps.