MOSS-Image Edit LoRA adds text-instruction image editing to MOSS-Image, with native ComfyUI support
MOSS Image Edit is a free community LoRA that lets the open MOSS-Image model edit an existing picture from a text instruction, changing colors, outfits, lighting, seasons or style while up to four reference images keep the rest. ComfyUI merged native support for it the same day, so it runs with stock loaders and samplers on your own GPU.
What it does
You connect one to four reference images and write what should change; with more than one image you name each image's role in the instruction. The rank-64 adapter (about 297 MB) works with either MOSS-Image-Base or MOSS-Image-SFT, and in core ComfyUI it loads through the normal LoRA loader plus a new, still experimental Text Encode MOSS-Image Edit node with edit_format set to Moss-Image-Edit.
Results and limits
The author's node pack reports test runs on 23 October with ComfyUI v0.42.3: the INT8 SFT model with an INT8 text encoder made single- and two-image edits on an 8 GB RTX 4060 Laptop, and BF16 made 1024×1024 two-image edits at 30 steps on an RTX 5090. The ComfyUI pull request also caches the reference images once, which took a 1024×1024 edit with a 1 MP reference from 826 ms to 457 ms per step.
Worth knowing
It is a community release by one author, not an official OpenMOSS one, and the card covers attribute and scene edits only, with no benchmark. Native support merged after ComfyUI v0.42.1, so update past that release; the author's ComfyUI-Moss-Image-Edit node pack and @LatentSpacer's native nodes (which also load a second, HQ edit LoRA) are other routes. The project code is GPLv3.
Try it yourself
Dress the person in image 1 in the outfit from image 2