# OOO_Anima v10
OOO_Anima v10 is an anime / illustration checkpoint fine-tuned from Anima Base v1.0 by CircleStone Labs / Comfy Org.
It was trained on approximately 53,000 images and is intended for anime-style, character-focused, and non-photorealistic illustration generation.
OOO_Anima v10 は、CircleStone Labs / Comfy Org の Anima Base v1.0 をベースに追加学習した、アニメ・イラスト向けCheckpointです。
約53,000枚の画像で学習しており、キャラクターイラストや二次元風・非写実的な表現に向いています。
## Recommended Settings
- Steps: 30-50
- CFG: 4-5
- Resolution: 512x512 ～ 1920x1920
- Samplers: euler_a,
## Prompt Example
Positive:
masterpiece,best quality, high quality,newest,year 2025, year 2024,safe,game cg,Negative:
worst quality, low quality, score_1, score_2, score_3, artifacts,early, old,nsfw,realistic, ## License
This model is a derivative of Anima Base v1.0.
It inherits the CircleStone Labs Non-Commercial License.
This model and its derivatives are for non-commercial use only.
Anima is also described as a derivative of Cosmos-Predict2-2B-Text2Image, so NVIDIA Open Model License Agreement terms may also apply as applicable.
Commercial use, selling this model, selling merges, or using it in paid generation services is not permitted.
## Credits
Base model: Anima Base v1.0 by CircleStone Labs / Comfy Org
Training tools: kohya / sd-scripts-style training tools
Description
FAQ
Comments (5)
Hi, nice model and great style/concept, but the model may be suffering from catastrophic forgetting.
Thank you! I’ll use the feedback as reference for future tuning.
If possible, could you share some specific examples of what made you feel that way?
@oron1208380 For example like some characters/artist have lower than 250 images in the danbooru the model can not generate correctly even though the base can generate them very easy
I tested the checkpoint.
And compared to others, yours handles 1920 x 1920 images very well.
I liked it.
it degrade outputs back to pony days