For best outputs:
Sampler: for real life outputs, Restart, for anime outputs, Euler a.
Steps: 20 for real life outputs, 30 for anime outputs.
Cfg 7, CLIP skip 2.
Pos prompt: for real life outputs, use realistic words like "Kodak" or "HD", the model is by default set to realistic. For anime outputs, i had my best results with "(Masterpiece:1.5), (best quality:1.5), detailed" plus the keyword "anime", sometimes being necessary to weight it.
Neg prompt: for real life outputs, nothing, or just what you dont want to see (nudity, nsfw, ect). For the anime outputs, "low-quality worst-quality:1.5), earrings, (bad-hands:1.35), multiple people, poorly drawn face, blur, blurry, nsfw, loli, kid, child, underage, topless, nipples, negative_hand-neg, ng_deepnegative_v1_75t, easynegative" being the embeddings "negative_hand-neg", "ng_deepnegative_v1_75t" and "easynegative" included in the anime example runs.
For best outputs:
Sampler: for real life outputs, Restart, for anime outputs, Euler a.
Steps: 20 for real life outputs, 30 for anime outputs.
Cfg 7, CLIP skip 2.
Pos prompt: for real life outputs, use realistic words like "Kodak" or "HD", the model is by default set to realistic. For anime outputs, i had my best results with "(Masterpiece:1.5), (best quality:1.5), detailed" plus the keyword "anime", sometimes being necessary to weight it.
Neg prompt: for real life outputs, nothing, or just what you dont want to see (nudity, nsfw, ect). For the anime outputs, "low-quality worst-quality:1.5), earrings, (bad-hands:1.35), multiple people, poorly drawn face, blur, blurry, nsfw, loli, kid, child, underage, topless, nipples, negative_hand-neg, ng_deepnegative_v1_75t, easynegative" being the embeddings "negative_hand-neg", "ng_deepnegative_v1_75t" and "easynegative" included in the anime example runs.
