Generate AI images locally on your own computer using a browser interface, with no coding required.
Run large AI image models on a GPU with limited VRAM using compressed NF4 or GGUF model formats.
Use ControlNet to generate images that match a specific pose or composition from a reference photo.
Install community-made extensions for new image generation styles and techniques.
Requires a CUDA-capable Nvidia GPU, the one-click installer automates Python and library setup but GPU hardware is mandatory.
Stable Diffusion WebUI Forge is a modified version of the popular Stable Diffusion WebUI, which is a browser-based interface for running AI image-generation models on your own computer. Forge is built on top of that existing interface and adds improvements to how it manages GPU memory, how fast it generates images, and how easily developers can add new features. The name is a nod to Minecraft Forge, which is a platform for adding mods to Minecraft. In the same spirit, Forge aims to be a stable foundation for building extensions and experimenting with new AI image techniques, while keeping compatibility with the large library of existing models, style presets (called LoRAs), and control extensions people have already built for the original WebUI. For someone who just wants to use it, Forge offers a one-click installation package: download a compressed file, extract it, and run a batch script to start. No manual Python or library setup is required. The interface runs locally in your browser. It supports a wide range of model formats, including compressed model formats (such as NF4 and GGUF variants) that let people with less GPU memory still run larger models. There is also a "GPU Weight" slider that controls how much of a model stays on the GPU at once versus being swapped to system RAM. Forge includes integrated support for ControlNet, which is a technique for giving the image generator more specific instructions about composition or pose. It also supports IP-Adapters and Instant-ID, which are methods for making generated images resemble a reference photograph. The README includes a status table listing which components are working, which are broken pending a fix, and which have not yet been implemented. Most core features were reported as working as of late 2024. There is also developer documentation showing how to write custom backend patches, called UnetPatchers, which let extension authors modify how the AI model processes images.
← lllyasviel on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.