ComfyUI Wiki · 工作流与开源创作分享·· 9 days agoAI score64
MiniMax-H3-Image integrates pDMD 2-NFE LoRA, completing sampling in 2 steps
Text to image
AI briefing
akhaliq merged the pDMD 2-NFE LoRA into the Comfy-Org checkpoint, reducing MiniMax-H3-Image sampling from 6 steps to 2 steps; a single 4096x4096 image on H200 went from 25.1 seconds to 17.4 seconds. The LoRA has been pre-merged into the 40GB weight file, or the 2GB LoRA can be loaded separately into an existing checkpoint; 2 steps work for most prompts, but graphic prompts are recommended at 3 steps.
Open original comfyui-wiki.com
MiniMax / HailuoImage GenerationComfyUI & WorkflowsTutorials & Practice#Tutorial#ComfyUI & workflows#Image generation#MiniMax