Luo Xiaohei style Lora

Adaptability to other models should be quite poor based on previous experience; it can only be paired with the recommended base model (the one used during training).

Therefore, you must exclusively use the model from the link:

Trigger word: luoxiaohei. The character is exclusively Luo Xiaohui in human form. Due to time constraints, nearly 90% of the images used were captured from the anime series. There was simply no extra time to capture or label images for other characters.

Sampling method: Euler a/Euler/DPM++2M

Iteration steps: 30

CFG: 7-8

Resolution: 1024×1024

All other settings default

Negative prompt examples: deformed hands, mutated fingers, deformed palms, deformed feet, polydactyly, multiple arms, fused fingers, polydactyly, low quality, blurry, borders

Note: No need to add quality prompts; use the provided negative prompts directly.

The model may omit parts not specified in the prompt, so detailed prompts are recommended. Natural language prompts are not supported.

Theoretically, both black and white hair should work, but white hair yields better results in practice.

For clothing, currently only white tank tops and black shorts are available; use "shoes" for footwear.

Default weight of 1 is sufficient.

In the anime, Luo Xiaohui's character rarely shows a full head or both ears simultaneously. Thus, full-body front-facing images are challenging, especially with ears requiring constant adjustments.

The background tends to be monotonous—just red sand dunes, blue sky, and forests. Since Xiaohui's human form appears in limited settings, pure white backgrounds are common.

NSFW: Not suggestion