Open-source project
YanWenKun/ComfyUI-Windows-Portable avatar
YanWenKun/ComfyUI-Windows-Portable

YanWenKun/ComfyUI-Windows-Portable: a preloaded ComfyUI pack for NVIDIA Windows machines

🎨ComfyUI standalone pack with 40+ custom nodes. | ComfyUI 大号整合包,预装大量自定义节点(不含SD模型)

610 stars112 forksPythonGPL-3.0

At a glance

What is it?
This is a Windows-only, NVIDIA-only ComfyUI distribution that ships 40+ custom nodes and 300+ Python packages already resolved. It removes dependency building from setup, and it also removes your control over what is inside the environment.
Who is it for?
Adopt this if you are on Windows with an NVIDIA GPU from 2018 or newer and you want the custom node ecosystem working without compiling insightface or dlib yourself. Do not adopt it if you are on AMD, on Linux, or if you need a clean dependency set you control.
Can I use it commercially?
Yes, with conditions. GPL-3.0 is a copyleft licence: if you distribute software that includes it, you must release that software's source code under the same licence. Running it internally without distributing it does not trigger that obligation.
Is it still maintained?
Yes. The repository last received commits 4 days ago.
What is it written in?
Mainly Python, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on September 18, 2026, and from our analysis. They are not legal advice.

DEEP OPEN-SOURCE ANALYSIS

What problem the YanWenKun pack removes, and who still has to do the work

Installing ComfyUI by hand is mostly a Python dependency problem. Custom nodes pull in packages that need native compilation, and insightface and dlib are the two the README names explicitly as examples of packages that normally require building during installation. On Windows that build step is where setups usually fail. This repository ships those packages already built, together with performance libraries SageAttention and xFormers, so the first launch does not begin with a compiler error.

The target user is narrow. The README states that only NVIDIA GPUs from 2018 or newer are supported, and it strongly recommends installing the latest driver. There is no CPU path and no AMD path documented. If you are on an AMD card, the search traffic around comfyui_windows_portable_amd.7z does not correspond to anything this repository offers; the download list here is cu130 only.

One thing the pack deliberately does not solve is models. The README is direct: only a few functional models are preinstalled, and users are told to prepare their own model files under ComfyUI\models\. So you get the runtime and the node library, not the weights. Budget the model download separately, and note the README's warning that nodes which fetch model files on first run can freeze the interface on a poor connection.

Split 7z archives, the models.zip overlay and the 260-character path limit

The distribution is a portable folder, not an installer. There is no registry entry and no system-wide Python. Everything lives under the directory you extract to, which is why the README's first instruction after downloading is about path length: extract to a shallow directory such as C:\AI\ to stay under the MAX_PATH limit of 260 characters.

The archive itself is split compression. You download three files, ComfyUI_Windows_portable_cu130.7z.001 through .003, and the README states that you open only the 001 file; the extraction tool handles the rest because they are parts of one archive. The README suggests 7-Zip. A separate models.zip carries the few functional models, and the README notes you can skip it when upgrading from a previous release.

This is the part of the design most likely to confuse a first-time user. Three files that look like independent downloads are actually one, and extracting 002 or 003 on its own produces nothing useful. The README addresses this directly, which suggests it is a recurring support question rather than a theoretical concern.

Installing ComfyUI Windows Portable from the release page and running it once

Download the three split archive parts plus models.zip from the release page, place them in one directory, and extract the 001 file there. The README gives direct links to the latest release assets, so the filenames below are the ones to expect:

bash
ComfyUI_Windows_portable_cu130.7z.001
ComfyUI_Windows_portable_cu130.7z.002
ComfyUI_Windows_portable_cu130.7z.003
models.zip

Extract the 001 file with 7-Zip into a shallow path. Then open models.zip and extract it to the same location, so the model folders land inside the portable tree. Your own model files go under the ComfyUI\models\ folder, per the README. Nothing else needs configuring before the first run.

Start the application by running the launcher batch file from the extracted folder:

bash
RUN_Launcher.bat

The README states that the program opens a browser automatically after launching, and that you can also reach the interface manually at http://localhost:8188/. Closing the window exits the program. If the browser does not open, the port is the thing to check first, since it is fixed in the documentation rather than configurable through a flag.

ComfyUI Manager is the update mechanism, and Update All is the risk

The pack ships ComfyUI-Manager, reachable through the Manager button in the top-right of the ComfyUI page. The README frames it as the way to install, update, disable and uninstall custom nodes, which means the preinstalled node list is a starting point rather than a fixed set.

The README draws a line between two update actions. Clicking Update ComfyUI is described as something you can do daily; it refreshes workflow templates and related updates without touching installed custom nodes. Update All is the one to treat carefully. The README advises against updating unused nodes, on the grounds that doing so risks Python package conflicts.

That advice is the honest admission at the center of this project. A preloaded environment with 300+ packages is a solved dependency graph at packaging time, and every node you update afterwards can move that graph. The pack's value comes from the frozen state, and the Manager button is the tool that unfreezes it. The README's own guidance, update what you use and leave the rest alone, is the correct reading.

Where this pack is the wrong tool: AMD, Linux and clean environments

The clearest limitation is hardware. The README states NVIDIA only, 2018 or newer, and the release assets are cu130 builds. There is no documented AMD or CPU fallback in this repository, so the AMD-related search terms people use do not map to anything offered here.

Platform is the second boundary. This is a Windows portable package. On Linux you would be better served by a normal ComfyUI install, since the whole point of the portable layout is avoiding Windows path and compilation problems that do not exist in the same form elsewhere.

The third case is subtler. If you need to know exactly which version of every package is present, a 300+ package environment that someone else resolved is a liability. You inherit their compatibility decisions, and when a node breaks you debug a dependency graph you did not build. For a production pipeline with pinned requirements, a manual install is more work upfront and less work later.

Finally, disk and network cost are real. The archive is split into three parts, there is a separate models.zip, and the README warns that first-run model downloads can freeze the interface on slow connections.

How it differs from the official ComfyUI portable build

The README positions this pack explicitly against the ComfyUI official standalone portable from Comfy-Org, describing it as similar but preloaded with numerous custom nodes and Python packages, with all dependencies resolved.

That is the whole difference, and it is a meaningful one. The official portable gives you ComfyUI and its core dependencies; you add nodes through the Manager and accept that some of them will need compilation. This pack gives you the node library already present, including the categories the README lists: performance nodes such as GGUF and RadialAttention, workspace tools such as Crystools and the ComfyUI-to-Python Extension, control nodes such as ControlNet Auxiliary Preprocessors and IPAdapter plus, and video nodes such as Video Helper Suite and AnimateDiff Evolved.

The trade-off is version lag. The official build tracks ComfyUI itself closely. This pack is assembled through a GitHub pipeline, so its contents reflect whenever that pipeline last ran. The releases listed are v11.1 from 2026-08-07, v11.0 from 2026-07-09 and v10.15 from 2026-05-25, and the last push to the repository was on 2026-09-14. If you need a node added in the last few weeks, install it yourself through the Manager rather than waiting for the next pack.

Licence, packaging pipeline and what an upgrade actually costs

The repository is GPL-3.0. That covers the packaging scripts and repository contents; the bundled custom nodes and Python packages carry their own licences, and the README does not enumerate them. If you redistribute a build, that is the area to check, and it is not something this article can settle.

The README also states that packaging runs through a GitHub pipeline, and that forking the repository is enough to run the packaging script without configuring CI/CD. That matters for two reasons. First, you can build your own all-in-one package with a different node set. Second, it means the published pack is reproducible from the repository rather than assembled by hand.

Upgrade cost is low by design. The README notes that when upgrading from a previous release you do not need to download models.zip again, so an upgrade is the archive plus your existing models folder. The practical cost is disk space for a second extraction, plus re-checking any nodes you installed or updated manually through the Manager, since those changes are not part of the packaged archive.

Editorial conclusion

Adopt this if you are on Windows with an NVIDIA GPU from 2018 or newer and you want the custom node ecosystem working without compiling insightface or dlib yourself. Do not adopt it if you are on AMD, on Linux, or if you need a clean dependency set you control. Before committing, confirm your free disk space and that your extraction path is shallow enough to stay under the 260-character MAX_PATH limit, since the README states long paths cause extraction errors.

Frequently asked questions

Is there a portable version of ComfyUI for Windows?

Yes. YanWenKun/ComfyUI-Windows-Portable is a Windows portable package that ships ComfyUI with 40+ custom nodes and 300+ Python packages preinstalled, and it is positioned in the README as similar to the official ComfyUI standalone portable but preloaded.

Should I install ComfyUI desktop or portable?

The README does not compare the two, so the choice depends on your constraints. This pack is a portable folder with no installer, requires an NVIDIA GPU from 2018 or newer, and is started with RUN_Launcher.bat.

Can you use ComfyUI offline?

The README notes that some preinstalled custom nodes download model files on the first run, which can cause freezing on a poor connection. It does not state that the pack works fully offline out of the box.

How do I install ComfyUI Windows Portable?

Download the three split archive parts and models.zip from the release page, place them in one directory, extract the 001 file with 7-Zip into a shallow path such as C:\AI\, then extract models.zip to the same place. The README states you do not need to open the 002 and 003 files.

How do I update ComfyUI Windows Portable?

The README describes clicking Update ComfyUI in ComfyUI-Manager as a daily-safe action that refreshes workflow templates without affecting installed custom nodes. It advises caution with Update All, since updating unused nodes risks Python package conflicts.

How do I run ComfyUI Windows Portable?

Run RUN_Launcher.bat from the extracted folder. The README states the program opens a browser automatically, and that you can also reach the interface at http://localhost:8188/. Closing the window exits the program.

Official sources

  1. Issues
  2. License: GPL-3.0
  3. README
  4. Releases
  5. YanWenKun/ComfyUI-Windows-Portable on GitHub
Community notes

Community notes