if your reference image for the video contains the face you want to use, is it still necessary to train a whole hunyuan model?

Any chance you can upload your workflow? Dropping the video in to comfy doesn't work

How to see the video prompt in comfy?

Can you share dataset and other details used to train this LoRA? I'm curious how many different photos are necessary to train LoRA of such quality.

As per other comments, your config file for training would be really helpful, I still cannot get good results with artifacts, frozen backgrounds etc.

Failing that, if you could share number of steps, LR, any other key information it would be much appreciated!