Can I continue finetune a DMD distilled model？

[muziyongshixin]

As the DMD distillation strategy does not guarantee the flow matching process, Can I finetune a distilled model using the flow matching loss?

I have tried to finetune a distilled 3-step wan2.1 video generation model using the standard flow matching loss, but the result is weird. The generated video just becomes blurry gradually.

I am just a beginner, hope someone can give some advice.