Open l-cr opened 2 months ago
I follow this post from HuggingFace: https://huggingface.co/stabilityai/sd-vae-ft-mse-original/blob/main/README.md
The quality of original model is already good, just because I tried with smaller image it generates some visible artifacts, thus I just used simple losses. If fine-tune on a total new image domain, using adversarial loss is needed, indeed.
Hi! Why not use adversarial loss?