AUTOMATIC1111
The Stable Diffusion web UI is a tool developed by AUTOMATIC1111 and hosted on GitHub. It provides users with the ability to contribute to ...
Last verified:
What is AUTOMATIC1111?
Stable Diffusion web UI (also known as AUTOMATIC1111, A1111, or SD WebUI) is a browser-based interface built on the Gradio library that allows users to generate and edit images using Stable Diffusion models. It provides text-to-image (txt2img) and image-to-image (img2img) generation capabilities, along with advanced features like inpainting, outpainting, prompt editing, and multiple upscaling options. The interface exposes all Stable Diffusion parameters through an intuitive web UI while also adding custom features not present in the base model.
Key features include support for over 20 samplers (DDIM, Euler, Euler a, DPM++ 2M Karras, UniPC), attention control for prompt weighting, LoRA and hypernetwork support for model fine-tuning, CLIP interrogator for reverse-engineering prompts from images, checkpoint merger for combining models, and extensive batch processing capabilities. The tool supports SD-XL (including the built-in refiner since version 1.6.0), Stable Diffusion 2.0/2.1, Alt-Diffusion, and various variation models. It also includes neural network upscalers like GFPGAN, CodeFormer for face restoration, RealESRGAN, ESRGAN, SwinIR, and LDSR.
This tool is designed for artists, designers, AI art enthusiasts, hobbyists, and professionals who want local control over AI image generation. It runs locally on users' machines on Windows, Linux, macOS (including Apple Silicon), and supports NVidia GPUs (recommended), AMD GPUs, Intel CPUs/GPUs, and even CPUs. With optimizations for low VRAM GPUs (4GB support, and reports of 2GB working), it caters to users with varying hardware capabilities. The open-source, community-driven development ensures regular updates and a vast ecosystem of extensions.
AUTOMATIC1111 pricing
Pricing model: Free
Free and open-source under AGPL-3.0 license. No paid tiers, subscription fees, or usage limits. Users download the repository via Git and run it locally on their own hardware. All features are available without cost. The only expenses are purchasing hardware (GPU, computer) and any model files/checkpoints users choose to download from external sources. Community extensions are also free. Online services like Google Colab offer alternative hosting but may have their own pricing.
AUTOMATIC1111 pros
- Open-source and free with no usage limitations or costs
- Runs locally ensuring complete privacy and control over generated images
- Supports NVidia, AMD, Intel GPUs plus Apple Silicon and CPU
- Over 20 sampling methods including DDIM, Euler, DPM++ 2M Karras, UniPC
- Native SD-XL support with built-in refiner inference since version 1.6.0
- LoRA, hypernetworks, and Textual Inversion embeddings support built-in
- Advanced inpainting with mask drawing, upload mask mode, and only-masked area option
- CLIP interrogator to extract prompts from existing images
- Checkpoint Merger tab to combine up to 3 checkpoints into one model
- X/Y/Z plot for creating 3D parameter grids with varying values
AUTOMATIC1111 cons
- Requires installing Python and Git manually before first use
- Steep learning curve with many parameters overwhelming for beginners
- Significantly GPU and CPU resource intensive during generation
- Cannot use machine for other tasks while generating due to GPU usage
- Large disk space required for checkpoints (can exceed 200GB) and LoRA files
- Frequent updates may introduce new bugs or disrupt existing functionality
- Support is primarily community-driven via forums rather than dedicated customer support
- May have compatibility issues across different operating systems or hardware configurations
- Some features like LoRA not supported with SD 2.0+ networks in WebUI
- Prompt editing syntax does not work with extra networks like LoRA
Frequently asked questions about AUTOMATIC1111
What hardware do I need to run Stable Diffusion web UI?
The recommended hardware is an NVidia GPU, though AMD GPUs, Intel CPUs/GPUs, Apple Silicon, and even CPUs are supported. The tool has optimizations for GPUs with as little as 4GB VRAM (with --lowvram or --medvram flags), and there are reports of it working with 2GB VRAM. You must have Python 3.10.6 (or 3.11 with configuration) and Git installed. For Windows 10/11 with NVidia, you can use the release package (sd.webui.zip) with batch files. Linux users need to install wget, git, python3, and relevant libraries via their package manager.
How do I install Stable Diffusion web UI on Windows?
For automatic installation on Windows: 1) Install Python 3.10.6 with 'Add Python to PATH' checked, 2) Install Git, 3) Clone the repository using 'git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git', 4) Run webui-user.bat from Windows Explorer as a normal user (not administrator). Alternatively, use the release package: download sd.webui.zip from v1.0.0-pre, extract it, run update.bat, then run run.bat.
What is the difference between txt2img and img2img modes?
txt2img (text-to-image) generates images from scratch based only on your text prompt. img2img (image-to-image) takes an existing input image and generates a new image based on both the prompt and the input image, allowing you to modify or transform existing images. img2img includes additional features like inpainting (draw a mask over parts to regenerate), outpainting (extend image beyond original boundaries), resizing options (just resize, crop and resize, resize and fill), and scripts like Loopback and Stable Diffusion Upscale.
How do I use LoRA networks in the web UI?
LoRA networks are supported natively since the WebUI includes built-in support. Place LoRA files (.pt or .safetensors) in the models/Lora directory. To use a LoRA, add '<lora:filename:multiplier>' to your prompt, where 'filename' is the LoRA file name without extension, and 'multiplier' is a number typically from 0 to 1 controlling strength. LoRA cannot be added to the negative prompt. Note that LoRA networks for Stable Diffusion 2.0+ models are not supported by the WebUI.
What is Hires. fix and how do I use it?
Hires. fix is a convenience option that partially renders your image at lower resolution, upscales it, then adds details at high resolution in a second pass. This avoids the horrible images SD1/2 models produce at very high resolutions since they were only trained at 512px or 768px. Enable it by checking the 'Hires. fix' checkbox on the txt2img page. You control the small picture resolution via width/height sliders, and the large picture dimensions via 'Scale by' multiplier, 'Resize width to', and 'Resize height to' sliders. The feature is equivalent to generating, upscaling, then running img2img refinement.
How does prompt editing work with the [from:to:when] syntax?
Prompt editing lets you change the prompt mid-generation. The syntax is [from:to:when] where 'from' and 'to' are texts and 'when' defines when to switch. If 'when' is 0.0-1.0, it's a fraction of steps; if it's an integer >0, it's the step number. Example: 'a landscape' starts drawing 'fantasy landscape', then after step 16 switches to 'cyberpunk landscape'. You can also use [to:when] to add text after steps, or [from::when] to remove text. Nesting works too. Note this syntax doesn't work with extra networks like LoRA.
What upscalers are available and how do I use them?
The Extras tab includes multiple neural network upscalers: GFPGAN and CodeFormer for face restoration, RealESRGAN and ESRGAN (with many third-party models) for general upscaling, SwinIR and Swin2SR, and LDSR for latent diffusion super resolution. Place ESRGAN models (.pth files) in the ESRGAN directory. You can also select upscalers in the Hires. fix dropdown, including latent space upscalers. For SD Upscale (img2img tab), select 'SD upscale' from scripts to upscale via tiles with img2img refinement.
How do I fix black images with SD 2.0 or 2.1 models?
SD 2.0 and 2.1 are more sensitive to FP16 numerical instability due to their new cross attention module. If generating black images, enable full precision with the '--no-half' command line argument, or try using the '--xformers' optimization. You can add these to COMMANDLINE_ARGS in webui-user.bat. For fp16 enablement, you can also set ATTN_PRECISION=fp16 in webui-user.bat as shown in the documentation.
What is the X/Y/Z plot feature and how do I use it?
X/Y/Z plot creates grids of images with varying parameters. X and Y serve as rows and columns, while Z is the batch dimension. Select parameter types in X type, Y type, and Z type fields, then input values separated by commas in the corresponding value fields. Ranges are supported: '1-5' gives 1,2,3,4,5; '1-5 (+2)' gives 1,3,5; '1-10 ' gives 1,3,5,7,10. Prompt S/R (search/replace) mode substitutes keywords. This is useful for comparing how different samplers, steps, CFG scales, or other parameters affect output.
Can I run Stable Diffusion web UI without a GPU?
Yes, the tool can run on CPUs, though generation will be significantly slower than on GPUs. Intel CPUs and GPUs (both integrated and discrete) are supported via an external wiki page with instructions. The tool also supports Ascend NPUs (external wiki page). For users without suitable hardware, online services like Google Colab offer alternatives, though Colab's free version lacks advanced functionality like inpainting, ControlNets, and Adetailer that the WebUI provides.