Closed liuzysy closed 1 year ago
If you want to use CLIP. You probably don't need to change much. Just making the model a bit bigger and then instead of conditioning on learned class embeddings as I do in the code, you just provide in the CLIP embedding. CLIP embeddings are also just a vector, so same as class embeddings. Does that help?
Does anyone have some experience on img2img by using diffusion model?