LoRA for generating in the style of doujinshi artist Anon 2-okunen.
MOSS
Trained on MOSS-Image-Base, tested on MOSS-Image-SFT
Examples made on default comfy workflow.
er_sde + simple, 30-40 steps, cfg 3, lora strength 1. Resolution - 2 mpx on multiple of 16.
Stating the obvious - to render in this style you need it to be illustration, 2d, anime, etc.
Tl:dr do not use it with realistic description if your goal is to emulate this style
Anima
Make sure to use keyword @Anon 2-okunen
Tested on turbo and hires loras and without them.
Tested on natural language, CLIP styled prompts and mix of them.
All of examples are made between 0.8 and 1.
Without turbo model add "text" tag in negatives.
Quality tags masterpiece, best quality, score_7 and such are important.
Illustrious
Make sure to use keyword anon2okunen
Model is prone to fill the scene with characters, so use 'solo', '2girls' and negatives 'multiple characters', 'multiple girls', etc
neg recommended: (comic, 2koma, speech bubble, text, thought_bubble, chinese_text, japanese_text, onomatopoeia)Works with strength between 0.7-1, 1 can be glitchy depending on complexity of the prompt.
All of examples are made in 1, you need to be specific about the scene.