city96/ComfyUI-GGUF
Source: https://github.com/city96/comfyui-gguf
Structured fields
{
"archived": false,
"checked_at": "2026-10-17T00:00:00.000Z",
"usefulness": "Loader nodes for GGUF-quantized DiT/transformer models (Flux.1, SD3.5) and quantized T5 text encoders in ComfyUI, to fit large models into low VRAM. Practical bridge between llama.cpp-style quantization and ComfyUI recipes.",
"source_type": "repository",
"availability": "available",
"dependencies": [
"gguf (pip install --upgrade gguf)",
"recent ComfyUI with custom-op support"
],
"next_check_at": "2026-10-31T00:00:00.000Z",
"known_limitations": [
"Fetched page did not show date of latest commit, so current maintenance is unconfirmed; 193 commits and 201 open issues."
],
"upstream_updated_at": "2026-01-12T13:02:18.000Z"
}Evidence claims
- [upstream_claim] README: adds GGUF quantization support; the 'GGUF Unet loader' (category bootleg) loads .gguf files from ComfyUI/models/unet; T5 quants use *CLIPLoader (gguf); pre-quantized links for Flux.1-dev/schnell and SD 3.5 Large/Large Turbo; LoRA loading described as experimental. source
- [upstream_claim] README calls the project 'very much WIP', says quantization was not feasible for conv2d UNET models but works better for transformer/DiT models, requires a ComfyUI recent enough for custom ops, and notes macOS Sequoia seems to need torch 2.4.1 (issue #107). source
- [upstream_claim] Research review verdict at export: reviewed (as of 2026-10-17). Recorded by the research pipeline, not a hub review outcome; evidence.review stays unreviewed until an independent reviewer records one.
Rights
| component | license | hosting | redistribution | modification | training | Evidence and notes |
|---|---|---|---|---|---|---|
| code | Apache-2.0 | permitted | permitted | permitted | unknown | Repository page lists Apache-2.0. |
| model_weights | unknown | unknown | unknown | unknown | unknown | GGUF files are converted from base models (e.g. FLUX.1 dev) and inherit their licenses. Model files these nodes load or download carry their own licenses; the node code license does not cover them. Not assessed here. |