I saw the FAQ mentioning merging custom face models with your models trained on styles. You recommended img2img for this, but you can actually continue training on your models. I load them into fast-dreambooth (notebook by TheLastBen), and then train it as a normal dreambooth model, as if I was using sd1.5 itself. Make sure not to "resume" training.
I saw the FAQ mentioning merging custom face models with your models trained on styles. You recommended img2img for this, but you can actually continue training on your models. I load them into fast-dreambooth (notebook by TheLastBen), and then train it as a normal dreambooth model, as if I was using sd1.5 itself. Make sure not to "resume" training.
This has really good results, examples below.
Render using mo-di diffusion:
Example instance image: