generated at
同じ元画像からのimg2imgとLoopbackで一貫性のある肖像画を生成する
CFGは0.7 ~ 0.8
これはなんでも良い
テンプレート画像(元画像)も比較的何でも良い
ただし、これで構図が決定される
その分トークンが節約できる
ループバック回数はちょうど良いところを見極める必要がある
人狼のような元画像から大きく離れるものを作りたい際はPrompt Editingを用いて、狼から人狼へと変化させる
$ (monstrous) ([wolf:werewolf:0.45]:1.2) wearing a suit
prompt
Arcane Diffusion
(black and white) digital portrait of _insert subject here_, 1920s fashion, chiaroscuro, detailed face, detailed clothes, cinematic lighting, reflections, rich shading, deep shadows, low-key lighting, high quality, by Guido Crepax and Tomer Hanuka and Yoji Shinkawa, [from:arcane style:3]
Negative
Negative prompt: (bad_prompt:0.8),comic panels, white outline, sketch, repetitive, cropped, bad illustration, kids drawing, long neck, totem pole, multiple jaws, (flat)
settings
46 - 48, Sampler: DPM++ 2M Karras, CFG scale: 12.5, Seed: {various}, Size: 512x768, Model hash: 318a302e, Denoising strength: 0.75, Denoising strength change factor: 1 (also --xformers).
txt2imgとimage2imageで作っていた際は1時間かかっていたものが15秒で済むようになった
その他promptまとめ