512x512分辨率70+白底全身图片训练，形象比较还原，多个视图角度，关键词的anime换成3d用来生成3d类型图片效果也不错。

推荐尺寸512x512。

Steps: 20-30

CFG scale: 7-13

权重 0.5-1

采样 DPM++ 2M Karras， Euler a，可能较好

提示词：gengar, anime ,red eyes,spiky hair,pointed ears,fat body,three fingers,three toes,purple body,no humans,red sclera pokemon (creature),cute, front view,white background, solo, simple background, full body, tongue out, teeth, colored sclera, open mouth, purple theme,  :<, looking at viewer, standing ,tail，

模型试了Nai3-CHIBI4.5Mix ，Golden fantasy(金色幻想) ， PrimeMix 等都不错，高分修复重绘不要太高，会变样，0.3左右会好些

Resolution: 512x512
Trained on 70+ full-body images with white backgrounds, achieving high character accuracy. Multiple viewing angles included. Replacing the keyword "anime" with "3D" works well for generating 3D-style images.

Recommended size: 512x512
Steps: 20-30
CFG scale: 7-13
Weight: 0.5-1
Samplers: DPM++ 2M Karras and Euler a perform best.

Prompt example:
gengar, anime, red eyes, spiky hair, pointed ears, fat body, three fingers, three toes, purple body, no humans, red sclera pokemon (creature), cute, front view, white background, solo, simple background, full body, tongue out, teeth, colored sclera, open mouth, purple theme, :<, looking at viewer, standing, tail

Recommended checkpoints:

• NAI3-CHIBI4.5Mix

• Golden Fantasy (金色幻想)

• PrimeMix

Notes:

• Avoid setting the Hires. fix denoising strength too high (recommended ~0.3) to prevent distortion.