>>105619068
so basically it's not really a distillation method, but they noticed that the training method works so well we can go for a few steps and it'll work well anyway, really interesting, does that mean that going for a full 50 steps training will make the model as good as veo 3? :^)