用戶LoRA
chibi
llustriousXL-SmoothFT-SPO versions are recommended! Pony is SO passé.There are two variants :OG = trained on unedited images, HARD to get rid of texts reliably with just prompts - especially with portraits. If you are planning to edit the t2i output anyway or are willing to generate a lot of batches, pick this one. Or maybe you can find a way to suppress texts on your own. idk.FX = More epochs + I removed texts from ~30% of the images, I tried my best but it could be slightly less faithful to the original style. Much easier to generate textless images. If you're unsure, just pick this one.FX recommended settings :lora strength : 0.6 ~ 1.5 depending on prompts and other loras you added on top of itprompt : chibi negative : text (higher weight if you are having trouble)