The Stable Diffusion web UI is a Gradio-based web interface for Stable Diffusion, written in Python, that enables local deployment of text-to-image and image-to-image generation capabilities.
The interface implements comprehensive image generation features including original txt2img and img2img modes, outpainting, inpainting, and color sketch functionality. Users can employ prompt matrices to explore parameter variations and leverage attention mechanisms to emphasize specific text portions using syntax like double parentheses or weighted notation. The loopback feature enables iterative img2img processing, while X/Y/Z plotting allows visualization of images generated with different parameters in three-dimensional space.
Advanced training capabilities include textual inversion for custom embeddings, hypernetworks, and LoRAs, with preprocessing options for image cropping, mirroring, and autotagging using BLIP or DeepDanbooru. The extras tab provides multiple face restoration and upscaling options including GFPGAN, CodeFormer, RealESRGAN, ESRGAN, SwinIR, Swin2SR, and LDSR. The system supports checkpoint merging, allowing users to combine up to three model checkpoints into a single file, and includes a checkpoint reloading feature for on-the-fly model switching.
The interface accommodates hardware constraints with reported support for 4GB video cards and lower, includes negative prompts for excluding unwanted elements, and provides seed resizing and variation generation for controlled image modifications. Prompt editing enables mid-generation prompt switching, while the highres fix option produces high-resolution output without typical distortions. The CLIP interrogator estimates prompts from existing images, and composable diffusion supports multiple weighted prompts using uppercase AND separators without token limits.
Installation follows platform-specific workflows with one-click scripts for Windows, Linux, and Apple Silicon, though Python and Git remain required dependencies. The project supports multiple GPU architectures including NVidia, AMD, and Intel processors, with alternative deployment options via online services like Google Colab. Settings allow customization of UI element defaults and reordering of interface components.
Custom scripts and community extensions extend functionality, with features like the history tab for image browsing and aesthetic gradients for style-specific generation available through the extension ecosystem.
