Skip to content
ComfyUI Wiki · 工作流与开源创作分享·· 23 hours agoPicksAI score68

ComfyUI Usage and Measured Comparison of the MOSS-Image-SFT 4-Step Distilled LoRA

Text to image

AI briefing

lvladikov's MOSS-Image-SFT 4-step distilled LoRA is trained with eDMD and LADD. In the author's tests, 22 1024² images took 26.6 seconds, compared with 387.3 seconds for the 50-step baseline, and fine texture averaged 1.22 times that of the 50-step teacher.

Why it matters

The author provides ComfyUI loading parameters and a measured comparison for the 4-step distilled LoRA, allowing direct reproduction and evaluation of its impact on texture and text rendering.

Open original comfyui-wiki.com