Learn Stable Diffusion web UI (A1111)
The form-based way to run Stable Diffusion locally: txt2img, img2img, hires fix, LoRAs and the PNG Info tab that reads every setting back.
What Stable Diffusion web UI is good at
- Quick local generation with a familiar form
- img2img and inpainting
- Hires. fix for larger, cleaner images
- PNG Info: every setting embedded in the file
Quickstart
- Load a checkpointPick the model at the top; match the resolution to its native size.
- Set sampler and stepsThe model card's recommendation, then 20 to 30 steps.
- Fix the seed to tuneChange one setting per run.
- Use PNG InfoDrop any saved PNG into the tab to read its prompt and settings and send them back to txt2img.
Settings cheat-sheet
| Setting | What it does | Start with |
|---|---|---|
| Checkpoint | Base model | One you know |
| Sampling method / steps | Denoising method and count | Recommended sampler, 20 to 30 |
| Width / height | Output size | Native size |
| CFG scale | Prompt adherence | 5 to 7 |
| Seed | Repeatability | Fixed while tuning |
| Hires. fix | Two-pass upscale | On for finals, denoise 0.3 to 0.5 |
| Denoising strength | img2img change amount | 0.4 to 0.6 |
Stable Diffusion web UI guides
Guides for Stable Diffusion web UI are on their way. The quickstart and cheat-sheet above are kept current in the meantime.
The five most-asked questions
What is A1111?
AUTOMATIC1111's Stable Diffusion web UI, a form-based local interface with txt2img, img2img, inpainting, extensions and scripts. Forge is a popular fork with the same layout.
How do I use a LoRA?
Place it in the LoRA folder and add it to the prompt as <lora:name:weight>, with the model's trigger words. Keep the weight below 1 to start.
What is hires. fix?
A two-pass method: generate at the native size, then upscale and refine with a chosen upscaler and denoising strength. It avoids the duplication artefacts of generating too large at once.
Where are my settings saved?
In the PNG's metadata: prompt, negative prompt, steps, sampler, CFG, seed, size and model hash. The PNG Info tab reads them back and can send them to txt2img.
A1111 or ComfyUI?
A1111 is quicker to learn and fine for most still-image work; ComfyUI gives full control and video models. Many people use both.