• Guide (External site): English | Japanese

• Hugging Face

A fine-tuned version of the MOSS-Image-SFT model that reduces noise for a clearer look and delivers vivid colors, especially for anime illustrations.

Comparison with MOSS-Image-SFT (Original)

MOSS-Image-SFT_clear | MOSS-Image-SFT (Original)

Compared with the original MOSS-Image-SFT on the right, MOSS-Image-SFT_clear produces cleaner, less noisy results with improved detail expression.

Saturation and contrast are also higher, resulting in clearer, more convincing illustrations.

Recommended Settings

• CFG scale: 3-4

• Sampler: euler

• Scheduler: normal

• Steps: 20-40 (around 30)

The ComfyUI workflow PNG can be downloaded from Hugging Face.

Tips (External site)

• Precision comparison (BF16 / Int8_ConvRot / FP8_svd_scaled / MXFP8 / NVFP4): English | Japanese

• MOSS-Image-SFT_clear: English | Japanese

License

• Apache-2.0