Open loadfred opened 7 months ago
Currently, FastSD does not support textual inversion, but it is possible to add textual inversion support since we are using diffusers. I haven't explored negative embedding with LCM.
For LCM to work with a negative prompt guidance scale should be greater than 1 which makes image generation slow and high RAM usage.
Is it possible to add embeddings (textual inversions)? Like FastNegativeV2, EasyNegative, BadDream, UnrealisticDream for the negative prompt. Does it need to be added to fastsdcpu? https://huggingface.co/docs/diffusers/using-diffusers/loading_adapters#textual-inversion