# ComfyUI-Windows-Portable: a packaging repo that ships a compiler's work as a download > ComfyUI-Windows-Portable is not a ComfyUI fork. It is a packaging repository that resolves 300+ Python packages, including the ones that need a native compiler, and publishes them as a three-volume archive named for its CUDA build. The trade is explicit: you get a working environment and you give up freedom to update it. **YanWenKun/ComfyUI-Windows-Portable** — 🎨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） - Repository: https://github.com/YanWenKun/ComfyUI-Windows-Portable - Stars: 615 · Forks: 113 - Language: Python - License: GPL-3.0 - Published: 2026-09-17 · Updated: 2026-09-17 · Language: en - Canonical page: https://hysenlabs.com/projects/yanwenkun-comfyui-windows-portable ## This is a packaging repository, and there is no ComfyUI source in it The root holds nine entries: dotfiles, a licence, two AsciiDoc READMEs in English and Chinese, `docs/` for the screenshots, `archived/`, and `builder/`. That last directory is the project. There is no ComfyUI source here because the aim is to resolve an environment, not to host the application. The README describes the repository as using the GitHub pipeline for packaging and says the point is to make your own all-in-one package easy: fork the repository and run the packaging script, with no CI/CD configuration to write. That is the sharpest thing about the project. Everything difficult about assembling a working ComfyUI install happens once, on somebody else's machine, and the result is committed to a release page. It also means the content of the bundle is defined by a list in this repository rather than by anything upstream. When a node breaks against a new ComfyUI, this pack decides whether it matters. ## Three split volumes, a models.zip, and a CUDA version in the filename The direct download links are four files: `ComfyUI_Windows_portable_cu130.7z.001`, `.002` and `.003`, plus `models.zip`. The README states that the three numbered files are parts of a single large archive created with split compression, so you open only the first one and the extraction tool picks up the other two. 7-Zip is suggested. The filename carries the CUDA build it was made against, `cu130` in this case. That is the only place the CUDA version is discoverable, and it matters when you compare packages, because two archives with different CUDA prefixes are not interchangeable. `models.zip` has its own rule: if you are upgrading from a previous release you do not need it, since you already have the models. The extraction path gets its own warning. Extract to a shallow directory such as `C:\AI\`, because a long path will hit the 260-character MAX_PATH limit and produce errors. That single instruction is the most common installation failure in this kind of package, and it costs nothing to follow. ## NVIDIA GPUs from 2018 onward, and the driver is the one thing you install Only NVIDIA GPUs from 2018 or newer are supported, and installing the latest driver is strongly recommended. That is the entire hardware requirement section, and it is short because everything else has been bundled. Once the archive is in place you run `RUN_Launcher.bat`, the program opens a browser automatically, and the address is `http://localhost:8188/` if you need to go there yourself. Closing the window exits the program. It is worth being clear about what the pack does and does not remove from your setup. Python, the interpreter, the compiled wheels, the custom nodes and a set of functional models all arrive in the archive. The GPU driver does not, because no archive can carry it. So the promise is not zero configuration, it is one configuration step: the driver. Models are explicitly left to you. Only a few functional models are pre-installed and the instruction is to bring your own model files, placed under `ComfyUI\models\`. ## 300+ prebuilt packages, including the ones that need a compiler The dependency claim is specific in a way most bundles are not. The pack pre-installs more than 300 mutually compatible Python packages, and it calls out two categories by name. The first is performance work, naming SageAttention and xFormers as examples. The second is the category that normally defeats a one-click install: packages that require native compilation during installation, with insightface and dlib given as the examples, which are described as already built and ready to use. That second category is the real product. A user who follows the standard ComfyUI instructions and hits a wheel that will not build is precisely the user this archive is for, and the fix is that somebody else already compiled it. The cost is the same fact stated from the other side. A resolved environment of more than 300 packages is a set of versions chosen on one machine at one moment, and it is the reason the tips warn against updating everything at once. ## The tips section warns you off Update All ComfyUI-Manager is pre-installed and is the supported way to manage custom nodes: install, update, disable and uninstall, from the Manager button in the top-right of the ComfyUI page. Two behaviours are worth separating. The Update ComfyUI button can be clicked daily to get the latest workflow templates and related updates, and it does not touch your installed custom nodes. That is the safe daily action. Update All is the one flagged as risky. The README says to be cautious with it, and explains why: there is no need to update nodes you do not use, and doing so invites Python package conflicts. That warning is the clearest statement of the design in the whole document. A pre-resolved bundle is a snapshot, and a rolling updater applied to a snapshot destroys the property that made it valuable. The advice is not to update what you are not using, which keeps the resolution intact. The Manager path is also the answer to anyone who wants a node this pack does not ship: install it through the Manager rather than merging a second dependency set in by hand. ## Forty-six named nodes across six categories, and one maintainer appears five times The custom node list is grouped into categories, and the visible entries can be counted. Performance has two, GGUF and RadialAttention. Workspace has seven, including ComfyUI Manager, Use Everywhere, Crystools, Lora Manager and an image saver and selector pair. General has ten, including WAS Node Suite, KJNodes, Comfyroll Studio, ComfyUI Essentials and rgthree. Control is the largest at sixteen, covering ControlNet preprocessors, Advanced ControlNet, Layer Style, the Impact Pack family, IPAdapter plus and RES4LYF. Video has six, including Video Helper Suite, Frame Interpolation, AnimateDiff Evolved and FizzNodes. A final category is truncated in the page you are reading. Counting maintainers across the visible list is more informative than counting nodes. ltdrdata appears five times, with ComfyUI Manager, WAS Node Suite, Impact Pack, Impact Subpack and Inspire Pack. kijai appears three times with KJNodes, Depth Anything V2 and Florence2. So a bundle that runs other people's code concentrates a lot of trust in a few prolific authors. The GPL-3.0 licence covers this repository; the licences of the forty-plus bundled projects are theirs. ## Releases are v10.15, v11.0 and v11.1, with the last commit after the last tag The three most recent releases are v11.1 dated 2026-08-07, v11.0 dated 2026-07-09 and v10.15 dated 2026-05-25. The numbers cross a major boundary between 10.15 and 11.0, and each release title repeats its own date. The last push to the repository was on 2026-09-28, which is after the newest tag, and the repository is not archived. So work has landed since v11.1 without a new release, and the archive you download is v11.1. Documentation lives in two languages as AsciiDoc files rather than Markdown, with the English one at the root and a Chinese translation beside it. The screenshot assets referenced for the interface and the launcher live in `docs/`. One gap is worth naming because it is a question people ask: there is no uninstall procedure anywhere in the documentation. For a package that extracts a large archive and installs a driver alongside it, that is the one instruction a reader most often wants and least often finds. ## Conclusion Use this pack if you are on Windows with an NVIDIA GPU from 2018 or newer and you would rather not compile anything yourself. Do not use it if you need a clean, reproducible dependency tree or an offline machine, since the first run downloads models and updating everything at once is flagged as risky. Before downloading, note the archive is pinned to a cu130 CUDA build, keep the extraction path short because of the 260-character limit, and read the Update All warning in the tips before you start changing nodes. ## FAQ ### Is there a portable version of ComfyUI for Windows? Yes, this repository is one. It is described as similar to the official ComfyUI standalone portable but preloaded with more than 40 commonly used custom nodes and more than 300 mutually compatible Python packages already built. It requires an NVIDIA GPU from 2018 or newer. ### how to install comfyui windows portable Put the three split archive parts and `models.zip` in the same directory, then extract only the `.001` file, since the other two are handled automatically. Extract to a shallow path such as `C:\AI\` to stay under the 260-character limit, extract `models.zip` to the same place, and put your model files under `ComfyUI\models\`. You can skip `models.zip` when upgrading from an earlier release. ### how to run comfyui windows portable Run `RUN_Launcher.bat` from the extracted package. A browser opens automatically, and the address is `http://localhost:8188/` if you need it. Closing the launcher window exits the program. ### how to update comfyui windows portable Use ComfyUI-Manager, reached from the Manager button in the top-right of the ComfyUI page, to install, update, disable and uninstall custom nodes. The Update ComfyUI button can be used daily for workflow templates and does not affect installed nodes, while Update All is flagged as risky because it can cause Python package conflicts. ## Sources - [Issues](https://github.com/YanWenKun/ComfyUI-Windows-Portable/issues) - [License: GPL-3.0](https://github.com/YanWenKun/ComfyUI-Windows-Portable/blob/main/LICENSE) - [README](https://github.com/YanWenKun/ComfyUI-Windows-Portable/blob/main/README.md) - [Releases](https://github.com/YanWenKun/ComfyUI-Windows-Portable/releases) - [YanWenKun/ComfyUI-Windows-Portable on GitHub](https://github.com/YanWenKun/ComfyUI-Windows-Portable) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/yanwenkun-comfyui-windows-portable