license: wtfpl | |
Activate each lora with the character's name (e.g. if you load the willow lora, use "willow" in your prompt). | |
I've used AbyssOrangeMix3 with very good results, other models should work prety well. | |
Best strength for AOM3 seems to be anywhere between 0.50 and 0.70 (0.60 gives consistently good results). Go lower to get some interesting takes on the characters. | |
Hands should be pretty okay, but the eyes are almost always slightly imperfect: you can ignore them (they usually look okay) or fire up GIMP to fix them if you're an autist like me. | |
The following prompts work pretty well: | |
Positive prompt: masterpiece, hires, ultra-detailed, beautiful digital illustration, 1girl, cute, beautiful, /character name/, /everything else/ | |
Negative prompt: nsfw, (worst quality, low quality:1.3), (depth of field, blurry:1.2), (greyscale, monochrome:1.1), 3D face, nose, cropped, lowres, text, jpeg artifacts, signature, watermark, username, blurry, artist name, trademark, watermark, title, (tan, child, infant, toddlers, chibi, sd character:1.1), multiple view, reference sheet, (multiple characters), |