Mikoshii’s Ultra Lightweight MOSS Workflow
An all-in-one, easy-to-use workflow for MOSS-Image in ComfyUI. Everything is controlled from a few on/off switches, so you don’t have to rewire anything.
Features
Single model or model merging: generate with one model, or blend two MOSS-Image models live with a simple ratio slider, and save the merge when you like it
Manual or automatic prompts: type your own prompt, or let LC Vision write a detailed prompt from any image you load
img2img: start from your own image with an adjustable strength
SeedVR2 upscaling: high-quality upscale with a before/after Image Comparer slider
LoRA support through the Power Lora Loader, with turbo LoRA settings ready (8 steps, CFG 1)
Resolution selector with megapixel and aspect ratio controls
Organized auto-saving: images save into
Mikoshii's Workflow/<model name>/with the date, time and seed in the file nameA built-in guide (English, 中文, 日本語) inside the workflow, with clickable download links for every model and the folder each one goes in
What it’s good for
Fast everyday generation with MOSS-Image and its fine-tunes
Testing and blending MOSS models without making a new file every time
Turning reference images into prompts, or remixing them with img2img
Making final high-resolution images in one click
Limitations
MOSS-Image models only. It won’t work with SDXL, Flux, Krea, Z-Image and so on without changes.
Merging works best with two bf16 models. fp8/int8 models can only go in the Model 1 slot, because ComfyUI 0.46.x can’t merge a compressed model in the Model 2 slot.
SeedVR2 upscaling and LC Vision use a lot of VRAM. It was built and tested on an RTX 4090 (24 GB). On smaller cards, lower the megapixels and turn the upscaler off if you experience stutters or slow downs, but 1mp should be fine.
Requires ComfyUI 0.42.0 or newer.
Getting started
Load the workflow, then run Manager → Install Missing Custom Nodes and restart.
Download the models listed in the LC Note guide on the left side of the workflow. Every link and folder is in there.
Pick a model, type a prompt, and press Run.
Feedback and images are very welcome! ❤️
Description
Base version.
