Main activation tags /// Main outfit (top to bottom)
seiichikinoAP, 1boy, anime coloring /// glasses, blue polo shirt, pants, sneakers Style control:
for the original anime TV show style use prompt: anime coloring
to avoid the style from the anime, remove anime coloring from positives and use other style tags
Recommended samplers:
- Euler Ancestral Scheduler: Simple CFG: 4.0 Steps: 30 Lora strength: 1.0
Prompt tips (green - positive, red - negative):
- chubby/fat - if your model doesn't generate him fat enough
- white shirt, red polo shirt, light purple polo shirt, grey newsboy cap, white and green striped shirt, blue undershirt, brown coat, cowboy hat, bandana around neck - additional canon clothes
- stubble, messy hair - messy version
- six fingers, four fingers - helps Anima with proper finger count
The LoRas should work with Anima checkpoints. Tested working well with the following:
AnimaYume v0.1 [Preview1]
AnimaYume v0.25 [Preview2]
AnimaYume v0.4 [Preview3]
AnimaYume v0.5 [Base 1.0]
Last image in the gallery can be imported into ComfyUI with a basic workflow:
No custom nodes required.
Description
Re-trained the same dataset for AnimaPreview2.
My thoughts on Preview2 model:
- better prompt adherence
- better overall quality when used with character Loras - don't need to use fine-tunes such as animeYume
- better text rendering - I was able to get correct text generated much quicker
Overall really happy with Preview2, its a direct upgrade over Preview1 and hopefully fine-tunes will push the quality even further.