We can put a face in Stable Diffusion using LoRA and dreambooth checkpoint models. But both require training a new model, which can be time-consuming. What if you can inject a face instantly at sampling without training?

This ComfyUI workflow copies the face of a person from input images. It can be used like a custom LoRA and checkpoint model, but no training is required.

It is fast and convenient!

It takes input images like these (images from the LoRA training dataset):

And you can generate images with any prompt.