Azuki Shinatsu / 志那都 アズキ | Maken-Ki! / マケン姫っ！
详情
下载文件 (2)
关于此版本
模型描述
Trained on Hassaku, 15 images. I utilized ze data from Chacha LoRa, I used anime images only. You may want to use the "full body" prompt, depends on the results you get.
Yotta Mix v5 - works perfectly so far, might be a bit tough to get five fingers instead of four, "thumbs" don't like to generate.
MeinaPastel v4: Works absolutely okay, even fingers generate correctly it seems.
GrahamAingPikas - ClearVAE 2.3: Looks good too. I think it might be my favourite, kinda close race with Yotta v5.
Galena REDUX: Good NSFW results
Other models I used with acceptable results: RuviStyle_v2.1/v2.0, Dark Sushi 2.5D 大颗寿司2.5D, Dark Sushi Mix 大颗寿司Mix - colorful (baked vae), Anyhentai_20, Koji v2.0/v2.1 and few more. I use 0.85 LoRa strength, 7 CFG with 23 sampling step "Euler A" for my testing.
I'm using a strong LoRa, "Add More Details - Detail Enhancer" ( https://pixai.art/model/1622983818871110869) for rich background, I'm not sure how good this LoRa is at generating rich sceneries or props by itself.
Thumbnail Model: MeinaPastel v4