This is a model with a Style LoRA merged into the Anima Base.

It may not be suitable as a base model for training. It is recommended to use Anima Base as the base for training.

Only the U-Net has been modified; the Text Encoder and LLM Adapter have not been changed.

※※※ This checkpoint is not an SDXL-based model. ※※※

Generation settings

• resolution

1024x1024, 896x1152, 1152x896, etc.

1535x1536, 1344x1728, 1728x1344, etc.

• steps

30～50

• CFG

3.0 ～ 5.0

Prompt

• Quality tag

masterpiece, best quality, highly detailed, highres, absurdres, newest

Since this model is merged with a style LoRA, the use of @artist tag and score_n tag —which significantly affect the visual style—is not recommended.

• Negative tag

worst quality, low quality, score_3, score_2, score_1, old, oldest

Including the number of fingers and toes in the prompt can help improve the results.

• Positive

5-fingers, 1-thumb, 5-toes

• Negative

6-fingers, 6-toes

Please refer to this as well.