Recommended one-click package
Use the bundled CUDA 12.1 and PyTorch 2.3.1 environment when you want the documented Windows setup with Git and Python already included.
webui_forge_cu121_torch231.7zDownload the 7z package Get the original Forge files from lllyasviel/stable-diffusion-webui-forge. For Windows with an NVIDIA GPU, start with the CUDA 12.1 and PyTorch 2.3.1 package recommended in the project README.
About 1.9 GB · Git and Python included · Model files are downloaded separately
The Windows archive is the shortest documented route. A manual repository install provides more control but requires you to manage the environment.
The filename identifies the bundled CUDA and PyTorch environment. It does not describe your installed NVIDIA driver or guarantee compatibility with every GPU.
| Package | Environment | When to use it |
|---|---|---|
Recommendedwebui_forge_cu121_torch231.7z | CUDA 12.1 · PyTorch 2.3.1 | Recommended in the project README. Start here for the documented Windows package route. |
Older environmentwebui_forge_cu121_torch21.7z | CUDA 12.1 · PyTorch 2.1 | The older packaged environment. Use it only when you have a specific compatibility reason. |
Compatibility warningwebui_forge_cu124_torch24.7z | CUDA 12.4 · PyTorch 2.4 | The README warns that MSVC may be broken and xformers may not work in this package. |
The download gives you the application environment. A complete first launch still needs extraction, an update, and a compatible model.
Use a 7z-compatible extractor and choose a writable path such as C:\AI\Forge. Do not run Forge from inside the archive or a protected system folder.
update.batThe project README makes this an explicit first step. Wait for the console to finish before continuing.
run.batKeep the console open. Forge runs locally and prints the browser address there; closing the console stops the UI.
Forge is the interface, not a checkpoint. Download the model files required by your chosen model family and place them in the correct folders.
Start from the project repository, then use the platform-specific route. The Windows one-click archive is not a universal installer.
The project does not require a repacked installer, download manager, or third-party mirror.
The GitHub address must contain lllyasviel/stable-diffusion-webui-forge. Similar names can point to forks or unrelated bundles.
The recommended archive is webui_forge_cu121_torch231.7z. A renamed executable or custom installer is not the same release asset.
Confirm the exact GitHub source before making any exception. The release metadata does not publish a checksum for these assets, so this site does not claim an independent hash verification.
Models have their own publishers, licenses, sizes, and file requirements. A third-party “all-in-one” bundle changes the source and support boundary.
Short answers to the questions that matter before you commit disk space or change your Python environment.
Use the lllyasviel/stable-diffusion-webui-forge repository on GitHub. This page sends Windows downloads to that repository’s release assets and does not mirror the files.
The project README recommends webui_forge_cu121_torch231.7z, which packages CUDA 12.1 and PyTorch 2.3.1. It is the default choice for the documented Windows and NVIDIA route.
No. The package provides the WebUI, Git, Python, and its runtime environment. Download a compatible checkpoint or other required model files separately, then place them in the correct Forge folders.
Not for the Windows one-click package; it includes Git and Python. A manual Git installation requires your own compatible Git and Python setup.
Forge is open-source software distributed under the AGPL-3.0 license. Models are separate downloads and can have different licenses or usage terms, so check each model before using it.
The package is a prepared environment and repository checkout. Running update.bat updates the Forge checkout as instructed by the project README; the word latest on the release page is not a guarantee that every bundled runtime component is newly published.
No. The published one-click archives are the documented Windows and NVIDIA route. Use the platform-specific repository guide for Linux, macOS, or AMD, and verify the compute backend before installing.
Extract the entire 7z archive to a writable folder, run update.bat, then run run.bat. Keep the console open while Forge is running and add a compatible model before generating an image.
Download the recommended Windows package now, or choose a manual repository installation for another environment.