#7979th of 120 in Image and videoFor image, video and design, image and design

comfyui-nunchaku

ComfyUI nodes for running 4-bit SVDQuant diffusion models with Nunchaku

Releases and problems feed (RSS)

Stars
2.9k
License
Apache-2.0
Last commit
Sep 2026
Last release
Jan 2026
Language
Python
Stars, last 29 weeks+125
29 Apr 2026: 2,809 stars6 May 2026: 2,816 stars13 May 2026: 2,823 stars20 May 2026: 2,829 stars27 May 2026: 2,833 stars3 Jun 2026: 2,839 stars10 Jun 2026: 2,849 stars17 Jun 2026: 2,856 stars24 Jun 2026: 2,860 stars1 Jul 2026: 2,865 stars8 Jul 2026: 2,868 stars15 Jul 2026: 2,873 stars22 Jul 2026: 2,876 stars29 Jul 2026: 2,878 stars5 Aug 2026: 2,881 stars12 Aug 2026: 2,886 stars19 Aug 2026: 2,892 stars26 Aug 2026: 2,895 stars2 Sept 2026: 2,895 stars9 Sept 2026: 2,902 stars16 Sept 2026: 2,906 stars23 Sept 2026: 2,907 stars30 Sept 2026: 2,911 stars7 Oct 2026: 2,913 stars14 Oct 2026: 2,919 stars21 Oct 2026: 2,922 stars28 Oct 2026: 2,925 stars4 Nov 2026: 2,929 stars11 Nov 2026: 2,930 stars18 Nov 2026: 2,934 stars

2,809 on 29 Apr 20262,934 on 18 Nov 2026

From GitHub's daily star counts.

Checked on GitHub.

Overview

ComfyUI plugin that loads 4-bit SVDQuant-quantized diffusion models through the Nunchaku inference engine. It covers FLUX.1, FLUX.1-Kontext-dev, Qwen-Image, Qwen-Image-Edit and Z-Image-Turbo, with LoRA, ControlNet-Union-Pro and a 4-bit T5 encoder. The README says async offloading can bring Qwen-Image transformer VRAM down to about 3 GiB.

Who it is for: ComfyUI users running FLUX or Qwen-Image models on limited VRAM

Strengths

  • Pre-quantized 4-bit models published on Hugging Face and ModelScope, with example workflows
  • Qwen-Image async offloading cuts transformer VRAM to as little as 3 GiB
  • Supports multi-LoRA, ControlNet-Union-Pro, FB Cache and FP16 attention
  • Wheel installer node, and DeepCompressor for quantizing custom models

Weaknesses

  • Needs the separate Nunchaku engine wheel installed alongside ComfyUI
  • Only works with models quantized for Nunchaku, not arbitrary checkpoints
  • Last tagged release is January 2026, while the main branch is newer
  • Supported models are limited to the families the maintainers have quantized

What it needs

  • GPU required
  • Needs ComfyUI, Nunchaku
  • Models: FLUX.1-dev, FLUX.1-Kontext-dev, Qwen-Image, Qwen-Image-Edit-2509, Z-Image-Turbo

What changed

From the release notes. No flag means the note does not say so, not that the upgrade is safe.

Known problems

Open now, reported in the last 90 days

Titles are the reporters' own words, taken from the project's GitHub issues and ranked by reactions. We have not tried to reproduce any of them.

Also in Image and video

See all 120
Also in Image and video
RankProjectScore
1frigateLocal NVR with realtime object detection for IP cameras37.8k stars+950 (+2.6%) in 30 daysMIT94 out of 100
2toonflow-appInfinite-canvas tool for AI short drama and comic video production22.4k stars+3,446 (+18%) in 30 daysMIT89 out of 100
3MoneyPrinterTurboGenerates short videos from a topic, with script, footage, voiceover and subtitles144.9k starsMIT88 out of 100
#4InvokeAIWeb UI for diffusion image generation with inpainting canvas and node workflows29.1k starsApache-2.088 out of 100
#5autoclipTurns a video link into short clips with covers and captions12.5k stars+2,045 (+20%) in 30 daysMIT88 out of 100