MOSS_LUC [ moss-image-checkpoint ]

Moss Image Luc sft int8 convrot

[ V.2 ]

More stable, more detailed, and better at following prompts compared to V1.

Note: Since V1 is currently not being displayed, I launched only V2.

While this V2 version slightly tones down the distinct style of the V1 LUC series, I believe it offers improved versatility.

An anime-focused MOSS-Image merge from my Lucy series, quantized to INT8 with convrot after merging. Best suited for anime and illustrated styles rather than photorealism.

Recommended settings

• Sampler: DPM++ SDE or DPM++ 2M SDE

• Steps: 8–10 (start with 8)

• CFG: 2–5 (start with 3)

• MOSS Scheduler: Shift, value 5

• DMD acceleration LoRA: lora_latest.safetensors, strength 0.7–0.8. I currently use 0.8; the LoRA author's recommendation is 0.7.

Suggested positive prompt

masterpiece, best quality, amazing quality,

Suggested negative prompt

worst quality, low quality, lowres, bad anatomy, bad feet, bad hands, jpeg artifacts, web address, censored, mosaic censoring,

MOSS-Image Character & Style Catalog

https://civitai.red/models/3013755/moss-image-character-and-style-catalog-prompt-library-and-browser

Looking for characters or styles to try with MOSS-Image? Check out my free catalog, featuring 33,937 character images and 221 style examples, with search, filters, and prompt copying.

These are suggested settings for this merge, not requirements. Feel free to adjust them to your taste. Photorealism is not this model's strength; if you try it, keep CFG around 2–3.

High Recommended scheduler extensions

For Forge Neo, enable MOSS Scheduler and select Shift:
https://github.com/ukr8b3g-cmyk/sd-webui-forge-neo-moss-scheduler

For ComfyUI, use MOSS Shift Scheduler:
https://github.com/ukr8b3g-cmyk/ComfyUI-MOSS-Shift-Scheduler

DMD acceleration LoRA:
https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd/blob/main/lora_latest.safetensors