Stable Diffusion WebUI, usually called AUTOMATIC1111 or A1111 after its author, is the program that made local image generation popular. It runs in your browser from a local server and gives you a text to image tab, an image to image tab, inpainting, upscaling and a settings page that exposes almost every option a model has. Its extension system brought ControlNet, regional prompting, Deforum animation and hundreds of other tools to home users.

A1111 is a good fit if you work mainly with Stable Diffusion 1.5 and SDXL checkpoints, like a form-style interface, and want to follow the huge library of guides and YouTube videos made for it. Development has slowed and it does not support Flux or the video models well; for those, Forge, SD.Next or ComfyUI are better.

无需安装 – 在云端运行 AI 模型

BitVector Prism 是最容易上手的方式:选择模型、输入提示词,即可在简洁的网页应用中生成,无需任何配置。BitVector 也提供 Discord 和网页版(SpyGlass)。

试用 BitVector

你需要什么

Windows 10 or 11 or Linux with an NVIDIA card: 4 GB VRAM works for SD 1.5, 8 GB is comfortable for SDXL. AMD cards work on Linux (ROCm) and on Windows through DirectML (slower). Apple Silicon Macs are supported through a manual install and are fine for SD 1.5 and slow but workable for SDXL. Allow 10 GB for the program and its Python packages, plus 2 to 7 GB per checkpoint.

如何获取并安装 Stable Diffusion WebUI (AUTOMATIC1111)

Windows: install Python 3.10.6 (tick Add Python to PATH) and Git, download the sd.webui.zip package from the GitHub releases page, extract it and run update.bat once, then run.bat.
Alternatively, clone the GitHub repository with Git and double click webui-user.bat. The first start downloads PyTorch and a default SD 1.5 model, which takes a while.
Linux: run the webui.sh script from the repository; it creates the virtual environment and installs the dependencies. Mac: follow the Apple Silicon page in the wiki (Homebrew, Python 3.10, then ./webui.sh).
Copy your checkpoints (.safetensors) into models/Stable-diffusion, LoRAs into models/Lora, embeddings into the embeddings folder and VAEs into models/VAE.
Open http://127.0.0.1:7860 in your browser when the console says Running on local URL. Install extensions from the Extensions tab (Available > Load from) and restart the UI afterwards.
Low VRAM? Edit webui-user.bat and add --medvram (8 GB or less) or --lowvram (4 GB) to COMMANDLINE_ARGS. Add --xformers on NVIDIA for faster generation.

请只从本页的官方链接下载。其他来源的安装包可能过时或被篡改。ModelTrends.ai 不托管任何文件。

无需安装 – 在云端运行 AI 模型

PirateDiffusion 仅在 Telegram 上提供,面向专业用户:数千个模型、LoRA 和工作流通过聊天命令驱动,固定价格套餐无限生成。

试用 PirateDiffusion

如何使用 Stable Diffusion WebUI (AUTOMATIC1111)

Choose a checkpoint in the dropdown at the top left. Wait for it to load; the first load of a new model takes longer.
In the txt2img tab, write your prompt and a negative prompt (for SD 1.5 and SDXL, a negative like "blurry, low quality, deformed" helps a lot).
Set the size (512x512 or 512x768 for SD 1.5, 1024x1024 or 832x1216 for SDXL), the sampling steps (20 to 30), the CFG scale (6 to 8) and a sampler such as DPM++ 2M Karras.
Press Generate. The result appears on the right and is saved in outputs/txt2img-images. Set the seed to -1 for a new random picture each time, or fix it to repeat a result.
To use a LoRA, open the Lora tab under the generate button and click one; it adds <lora:name:1> to your prompt. Lower the number to reduce its effect, and add the trigger words listed on its model page.
Send a result to img2img or Inpaint with the buttons under the image to change parts of it, or enable Hires. fix to render a larger, sharper version.

技巧与常见问题

Missing VAE results in washed out colours with some SD 1.5 models. Download the standard vae-ft-mse VAE and select it in Settings > VAE, or set it to Automatic.
If the UI will not start after an update, delete the venv folder and start again; it rebuilds the Python environment.
Use the PNG Info tab to read the prompt and settings out of any image made by A1111 and send them straight back to txt2img.
Keep an eye on the console window: errors and download progress show there, not in the browser.
SDXL needs 1024 pixel sizes to look right; 512 pixel sizes give poor results with SDXL checkpoints.

可在 Stable Diffusion WebUI (AUTOMATIC1111) 中运行的模型

Stable Diffusion WebUI is strongest with Stable Diffusion 1.5 and SDXL checkpoints, their LoRAs and textual inversion embeddings, which is most of this catalog. Pony, Illustrious and other SDXL variants work as normal SDXL models. Flux and video models are not a good fit here.

Stable Diffusion 1.5 模型

Stable Diffusion XL 1.0 模型

无需安装 – 在云端运行 AI 模型

不需要 GPU、下载或 Python。这些服务在自己的服务器上运行同样的开源模型，你可以在一分钟内从浏览器、Discord 或 Telegram 开始生成。

试用 BitVector

试用 PirateDiffusion

链接

这些链接指向开发者自己的网站。ModelTrends.ai 与他们没有关联，也不托管任何软件。

其他本地应用

如何安装 ComfyUI

The node-based powerhouse for Stable Diffusion, Flux, Wan video and almost every new open source model. Free, open source, for Windows, Mac and Linux.