AUTOMATIC1111 Stable Diffusion web UI: Local AI Art
/stable-diffusion-webui
AUTOMATIC1111's Stable Diffusion web UI is a self-hosted Gradio interface for running Stable Diffusion on your own GPU. It bundles txt2img, img2img, inpainting, LoRA training, and a community extension system into one settings-driven interface instead of separate scripts. Reach for it if you want most of the Stable Diffusion tooling ecosystem wired into a single UI; skip it if a lighter, node-based workflow suits you better.
What is AUTOMATIC1111 Stable Diffusion web UI?
AUTOMATIC1111 Stable Diffusion web UI is a browser-based interface for Stable Diffusion, built with the Gradio library and released under the AGPL-3.0 license. It runs on your own hardware and puts text-to-image, image-to-image, inpainting, outpainting, and upscaling behind tabs in one page, alongside checkpoint management, an API, and a scripting system for community extensions.
Core Features for AI Art Generation
- ✓txt2img and img2img generation modes, plus outpainting, inpainting, and Color Sketch for extending or editing an image in place.
- ✓Attention syntax lets you weight parts of a prompt directly in the text, like `(tuxedo:1.21)`, and Ctrl+Up/Ctrl+Down adjusts it without retyping.
- ✓X/Y/Z plot renders a grid of images across up to three varying parameters, useful for comparing samplers or CFG values side by side.
- ✓Textual Inversion, Hypernetwork, and LoRA training and loading, with a dedicated tab to preview embeddings before adding them to a prompt.
- ✓An Extras tab runs GFPGAN, CodeFormer, RealESRGAN, ESRGAN, SwinIR, Swin2SR, and LDSR for face restoration and upscaling.
- ✓Checkpoint Merger blends up to three checkpoints into one, and checkpoints reload on the fly without restarting the server.
- ✓Generation parameters get embedded in the output PNG or EXIF metadata, so dragging a past image into the PNG Info tab restores the settings that produced it.
- ✓A built-in API and a custom-scripts system let community extensions add features without patching the core app.
Setting Up the Web UI Locally
On Windows with an NVIDIA GPU, the fastest path is the packaged release: download sd.webui.zip from the v1.0.0-pre release, extract it, run update.bat once, then run.bat to launch. For a source install on Windows, install Python 3.10.6 with the 'Add Python to PATH' option checked, since newer Python releases don't support torch, install git, clone the repository, then run webui-user.bat as a regular user. On Linux, install the system packages your distro needs (wget, git, python3, and a few graphics or tcmalloc libraries depending on the package manager), clone the repo or fetch webui.sh directly, and run webui.sh; webui-user.sh holds the launch options. Very new distros may need Python 3.11 or 3.10 installed separately, which the README calls out specifically for Ubuntu 24.04. AMD GPUs, Intel CPUs and GPUs, Ascend NPUs, and Apple Silicon each get a dedicated setup page linked from the README instead of inline steps, and Google Colab is listed among the online-service options if you'd rather skip local setup.
Strengths
- ✓One interface covers generation, inpainting and outpainting, upscaling, and training instead of stitching together separate tools.
- ✓Attention syntax and Prompt Editing give fine control over a prompt without switching tools mid-generation.
- ✓Generation parameters save into the image file itself, so a past render's exact settings are recoverable from the image alone.
- ✓The extension system and custom-scripts wiki mean a missing feature is often already solved by the community.
- ✓Hardware reach beyond NVIDIA: AMD, Intel, and Ascend NPU setups are documented, plus a packaged release for people who don't want to manage Python.
Hardware and Software Considerations
- △Python version matters here: 3.10.6 is what the README recommends, and newer Python releases don't support torch, which trips up anyone who installs whatever Python their OS shipped.
- △AMD, Intel, and Ascend NPU support routes through separate, externally hosted wiki pages rather than the main README, so those setup steps aren't part of the core documentation.
- △The repository's own GitHub topics list includes 'unstable' as a tag.
- △4GB of VRAM is the stated floor, and the 2GB figure is only user reports, not an official guarantee, so low-end cards are a gamble.
- △Executing arbitrary Python code through the UI requires enabling the --allow-code flag, an easy setting to leave on longer than intended on a shared machine.
- △The README defers most user-facing documentation to the project wiki, so day-to-day workflow details and troubleshooting live outside this repo.
Other Stable Diffusion Interfaces
Frequently Asked Questions
Stable Diffusion web UI states a 4GB video card as its supported floor, with some user reports of 2GB working. You'll also need Python 3.10.6 and git installed, since newer Python releases don't support torch, and the README recommends an NVIDIA GPU, though AMD, Intel, and Ascend NPU setups are documented separately.
Stable Diffusion web UI is released under the AGPL-3.0 license, per its GitHub repository listing.
AGPL-3.0 does permit commercial use, but it's a copyleft license: if you modify Stable Diffusion web UI and offer it over a network, you're required to release those modifications under the same license. Licenses for individual borrowed components are also listed in the app's own Settings -> Licenses screen, worth checking before shipping a commercial product.
Stable Diffusion web UI supports NVIDIA GPUs as the recommended path, plus AMD GPUs, Intel CPUs and GPUs (integrated and discrete), and Ascend NPUs, each pointing to its own setup page from the README.
For the packaged Windows release, Stable Diffusion web UI updates by running update.bat and then run.bat, per the README. Update steps for the git-clone installation methods aren't clearly documented in the README itself.
Stable Diffusion web UI recommends Python 3.10.6, since the README notes newer Python releases don't support torch. On newer Linux distributions, it also notes you may need to install Python 3.11 or 3.10 separately alongside the system default.
The problem it solves
Running Stable Diffusion from research code usually means separately wiring up a sampler, an upscaler, a face-restoration model, and some way to track which settings produced which image. Stable Diffusion web UI puts all of that behind one Gradio page: the same interface handles txt2img, img2img, inpainting, upscaling, and checkpoint management, and it writes the generation parameters into the output file itself, so you're not logging seeds and CFG values in a separate spreadsheet.
Best use cases
- •Generating images from a text prompt on hardware you control, without per-call API fees.
- •Iterating on a single image through img2img, inpainting, and outpainting instead of restarting from a fresh prompt each time.
- •Training small custom embeddings, hypernetworks, or LoRAs on a specific character or style once you have local GPU access.
- •Batch-testing prompt or sampler variations with the X/Y/Z plot before committing to a final render.
- •Automating image generation from other software through the built-in API instead of clicking through the UI each time.
Who should try it — and who should skip
Try Stable Diffusion web UI if you want one local interface that already wraps inpainting, upscaling, LoRA training, and prompt-weighting syntax, and you're comfortable running Python and git yourself. Skip it if you want a managed, no-install product, or if a node-based workflow like ComfyUI fits how you think about image pipelines better than a tab-per-feature layout.
Related repositories
Want a second opinion on stable-diffusion-webui?
Ask an AI that can read this page — one click and you get its take on stable-diffusion-webui.