note · kongo_jun · 图像视频创作分享·· 2026-10-19AI score40
Hands-on test of MOSS-Image's 8/4/2-step acceleration LoRA in ComfyUI
ComfyUIでMOSS-Imageの8/4/2step LoRAを試してみた
AI briefing
Hands-on test in ComfyUI of MOSS-Image sft's 8/4/2-step acceleration LoRA: on an RTX 4090, 8-step generation took 10 seconds, 4-step 9.5 seconds, and 2-step 9 seconds, all with 16GB VRAM. After the model loads, a single image can be produced in about 1-2 seconds, and reducing steps still generates some text. The LoRA is relatively large and requires about 1GB of extra VRAM.
Open original note.com
Image GenerationComfyUI & WorkflowsTutorials & Practice#Tutorial#Image generation#ComfyUI & workflows