登录

ComfyUI-nunchaku

nunchux-aiApache-2.0活跃维护

把 Nunchaku 的 4-bit 量化扩散模型推理接进 ComfyUI，让 FLUX 等模型在低显存下也能快速出图。

ComfyUI Plugin of Nunchaku

v1.2.1发布于 2026-01-26 18:53 UTC2,931star171fork同步于 2026-11-07 07:30 UTC
量化推理扩散模型低显存ComfyUI插件4-bit量化
对比 GitHub 官网comfyuidiffusionfluxgenaimlsys

怎么上手

由 README 提炼，执行前请确认
原文 
README 未给出安装命令，仅链接到安装文档 https://nunchaku.tech/docs/ComfyUI-nunchaku/get_started/installation.html
硬件要求按 README 说明

README 未给出具体显存门槛，仅提到 Qwen-Image 异步 offload 可把 Transformer 显存压到约 3 GiB，并支持 20 系 GPU 的 INT4。

下载：源码 zip·GitHub Releases（安装包）本站不镜像安装包

项目说明

由 README 自动翻译提炼
没人工点评？写下你的经验

ComfyUI 的 Nunchaku 插件，用 SVDQuant 4-bit 量化加速扩散模型推理

把 Nunchaku 的 4-bit 量化扩散模型推理接进 ComfyUI，让 FLUX 等模型在低显存下也能快速出图。

核心能力

  • 接入 Nunchaku 引擎运行 SVDQuant 4-bit 量化模型
  • 支持 FLUX.1、Qwen-Image、Z-Image 等量化模型工作流
  • 提供 NunchakuWheelInstaller 节点安装对应 wheel
  • 支持多 LoRA、ControlNet 与异步 offload 降低显存
  • 兼容 20 系 GPU 的 INT4 推理

容易踩的坑

需单独安装与平台匹配的 Nunchaku wheel，可用 NunchakuWheelInstaller 节点处理；模型需从 Hugging Face 或 ModelScope 另行下载 4-bit 量化版本；具体安装步骤与依赖版本需查阅官方文档。

内容由模型根据 README 生成，可能不完整或不准确，以官方文档为准。看原文 README

最近发版

5 个版本v1.2.1
全部 release
  1. v1.2.1最新正式版2026-01-26 · 2026-01-26 18:53 UTC

    What's Changed

    Full Changelog: https://github.com/nunchaku-ai/ComfyUI-nunchaku/compare/v1.2.0...v1.2.1

    What's Changed

    Full Changelog: https://github.com/nunchaku-ai/ComfyUI-nunchaku/compare/v1.2.0...v1.2.1

  2. v1.2.02026-01-12 · 2026-01-12 09:02 UTC
    查看更新说明

    What's Changed

    New Contributors

    Full Changelog: https://github.com/nunchaku-ai/ComfyUI-nunchaku/compare/v1.1.0...v1.2.0

  3. v1.1.02025-12-27 · 2025-12-27 06:46 UTC
    查看更新说明

    What's Changed

    Full Changelog: https://github.com/nunchaku-tech/ComfyUI-nunchaku/compare/v1.0.2...v1.1.0

  4. v1.0.2ComfyUI-nunchaku v1.0.22025-11-02 · 2025-11-02 04:55 UTC
    查看更新说明

    What's Changed

    New Contributors

    Full Changelog: https://github.com/nunchaku-tech/ComfyUI-nunchaku/compare/v1.0.1...v1.0.2

  5. v1.0.1ComfyUI-nunchaku v1.0.12025-09-14 · 2025-09-14 05:34 UTC
    查看更新说明

    What's Changed

    New Contributors

    Full Changelog: https://github.com/nunchaku-tech/ComfyUI-nunchaku/compare/v1.0.0...v1.0.1

用户点评

登录用户的真实使用记录，可按版本和硬件筛选
共 0 条

    还没有人点评。第一个说出真实体验的人最有价值。

    README 原文

    未翻译
    在 GitHub 打开

    This repository provides the ComfyUI plugin for Nunchaku, an efficient inference engine for 4-bit neural networks quantized with SVDQuant. For the quantization library, check out DeepCompressor.

    Join our user groups on Discord and WeChat for discussions—details here. If you have any questions, run into issues, or are interested in contributing, feel free to share your thoughts with us!

    Nunchaku ComfyUI Plugin

    comfyui

    News

    • [2026-01-12] 🚀 v1.2.0 Released! Enjoy a 20–30% Z-Image performance boost, seamless LoRA support with native ComfyUI nodes, and INT4 support for 20-series GPUs!
    • [2025-12-26] 🚀 v1.1.0: Support 4-bit Tongyi-MAI/Z-Image-Turbo! Download on Hugging Face or ModelScope, and try it with this workflow!
    • [2025-09-24] 🔥 Released 4-bit 4/8-step Qwen-Image-Edit-2509 lightning models at Hugging Face! Try them out with this workflow!
    • [2025-09-24] 🔥 Released 4-bit Qwen-Image-Edit-2509! Models are available on Hugging Face. Try them out with this workflow!
    • [2025-09-09] 🔥 Released 4-bit Qwen-Image-Edit together with the 4/8-step Lightning variants! Models are available on Hugging Face. Try them out with this workflow!
    • [2025-09-04] 🚀 Official release of Nunchaku v1.0.0! Qwen-Image now supports asynchronous offloading, cutting Transformer VRAM usage to as little as 3 GiB with no performance loss. You can also try our pre-quantized 4/8-step Qwen-Image-Lightning models on Hugging Face or ModelScope.
    • [2025-08-23] 🚀 v1.0.0 adds support for Qwen-Image! Check this workflow to get started. LoRA support is coming soon.
    • [2025-07-17] 📘 The official ComfyUI-nunchaku documentation is now live! Explore comprehensive guides and resources to help you get started.
    • [2025-06-29] 🔥 v0.3.3 now supports FLUX.1-Kontext-dev! Download the quantized model from Hugging Face or ModelScope and use this workflow to get started.
    • [2025-06-11] Starting from v0.3.2, you can now easily install or update the Nunchaku wheel using this workflow!
    • [2025-06-07] 🚀 Release Patch v0.3.1! We bring back FB Cache support and fix 4-bit text encoder loading. PuLID nodes are now optional and won’t interfere with other nodes. We've also added a NunchakuWheelInstaller node to help you install the correct Nunchaku wheel.
    • [2025-06-01] 🚀 Release v0.3.0! This update adds support for multiple-batch inference, ControlNet-Union-Pro 2.0 and initial integration of PuLID. You can now load Nunchaku FLUX models as a single file, and our upgraded 4-bit T5 encoder now matches FP8 T5 in quality!
    • [2025-04-16] 🎥 Released tutorial videos in both English and Chinese to assist installation and usage.
    • [2025-04-09] 📢 Published the April roadmap and an FAQ to help the community get started and stay up to date with Nunchaku’s development.
    • [2025-04-05] 🚀 Release v0.2.0! This release introduces multi-LoRA and ControlNet support, with enhanced performance using FP16 attention and First-Block Cache. We've also added 20-series GPU compatibility and official workflows for FLUX.1-redux!

    Getting Started

    Star History

    Star History Chart

    相关项目

    同分类里 star 最高的项目
    对比全部 →
    headroom
    headroomlabs-ai·Python

    在工具输出、日志、文件和 RAG 片段进入 LLM 前先做压缩，编码 Agent 省约 20% token，JSON 省 60-95%，答案不变。提供库、代理和 MCP 服务器三种用法。

    上下文压缩+4量化进阶
    79325+2806.2kv0.44.02026-11-11 04:46 UTC
    llmfit
    AlexsJones·Rust

    一条命令就能从数百个模型和供应商里筛出适合你硬件配置的本地大模型，省去逐个试跑和查显存的麻烦。

    本地推理+4量化入门
    38904+1772.5kv1.1.202026-11-10 10:03 UTC
    PaddleNLP
    PaddlePaddle·Python

    基于飞桨的 NLP 与大模型开发库，提供丰富模型库，覆盖 BERT、ERNIE、LLaMA 等，支持分布式训练、压缩和文档智能等任务。

    大模型开发库+4量化硬核
    12987+13.0krl-v1.0.02026-05-23 02:54 UTC
    bitsandbytes
    bitsandbytes-foundation·Python

    让大模型在普通显卡上跑起来的 PyTorch 量化库，支持 8-bit 和 4-bit 权重压缩，也是 QLoRA 微调背后的核心依赖。

    量化+4量化
    85379450.50.22026-11-02 14:36 UTC