search

yardenfren1996 / B-LoRA

Implicit Style-Content Separation using B-LoRA
MIT License
302 stars 22 forks source link

Training content only blocks do not work for person #1

Open jetjodh opened 8 months ago

jetjodh commented 8 months ago

I tried your code to train a blora model on a small set of images of myself but was not able to replicate my facial features at all when trained with content only blocks. Any suggestions?

yardenfren1996 commented 7 months ago

Hi, thank you for reaching out. Actually, it's a very interesting point you've made. I'm currently investigating that issue, and I hope to provide a satisfactory answer as soon as possible.

Thank you!

yardenfren1996 commented 7 months ago

@jetjodh By the way, note that during training our method on a set of images (instead of a single image), we discovered that using the explicit prompt "A [v] <class-name>" yields better results. I recommend you give that a try.

jetjodh commented 7 months ago

I'll try adding the class name to the prompt and try again.