Automatic1111 Fix, Crash & Optimization Guide

AUTOMATIC1111 Stable Diffusion WebUI is a popular browser-based interface for running Stable Diffusion models locally. It is used by artists and hobbyists for text-to-image, inpainting, and image-to-image generation.

Install / First Setup

Clone the repository and run the launcher for your OS:

git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
cd stable-diffusion-webui
# Windows
webui-user.bat
# Linux / macOS
./webui.sh

The first launch installs dependencies and opens http://127.0.0.1:7860. Place model checkpoints (.ckpt/.safetensors) in the models/Stable-diffusion folder. A CUDA-capable GPU with current drivers is required for GPU acceleration on Windows/Linux.

Common Issues & Fixes

torch not compiled with CUDA enabled / running on CPU

Cause: A CPU-only PyTorch was installed, or no compatible GPU/driver. Fix: Verify with python -c "import torch; print(torch.cuda.is_available())" from the webui Python. If False, the launcher will reinstall a CUDA torch when you set the appropriate COMMANDLINE_ARGS, or reinstall the matching PyTorch wheel manually.

Out of memory (CUDA OOM)

Cause: Model plus batch/resolution exceeds VRAM. Fix: Add memory flags in webui-user.bat (Windows) or the launch script:

set COMMANDLINE_ARGS=--lowvram --opt-sdp-attention

--medvram is less aggressive; --lowvram fits very small GPUs. Lower resolution and batch size, and disable extras like face restoration if OOM occurs.

Black or green images

Cause: Missing/incompatible VAE, or a silent VAE decode OOM. Fix: Select a matching VAE in Settings → VAE, or set the VAE path. If only large outputs are black, enable tiled VAE in the UI to avoid VRAM spikes.

xformers import error

Cause: xformers version mismatches the installed PyTorch/CUDA. Fix: Either remove --xformers from COMMANDLINE_ARGS, or reinstall the xformers build matching your PyTorch version. Using --opt-sdp-attention avoids xformers entirely on supported setups.

Paths with spaces / non-ASCII cause failures

Cause: The webui folder is under a path containing spaces or special characters. Fix: Move stable-diffusion-webui to a simple path such as C:\sd\ (Windows) or ~/sd/ (Linux/macOS).

Performance & Optimization

• Low-End (8 GB RAM, no/weak GPU): Use --lowvram, SD1.5 models (1–2 GB), resolution 512×512, batch 1. Avoid SDXL.
• Mid-Range (16 GB RAM, 6–8 GB VRAM): --medvram, SDXL at ~1024×1024. Use --opt-sdp-attention or --xformers for speed.
• Workstation (12–24 GB VRAM): Run without VRAM flags; fp16 weights halve model memory versus fp32; raise batch size and resolution. Use --opt-channelslast where beneficial.
• Keep the models and outputs folders on fast storage; large checkpoint loads are I/O bound.

Version & Compatibility Notes

The webui targets Python 3.10.6 (recommended) and PyTorch 2.x with CUDA on NVIDIA GPUs; AMD on Linux uses ROCm builds. xformers is optional and version-sensitive. For exact supported Python/PyTorch/CUDA combinations, consult the official AUTOMATIC1111 wiki and release notes.

FAQ

Q: Where do I put model files? A: Place .safetensors/.ckpt checkpoints in models/Stable-diffusion and VAEs in models/VAE; restart the UI to load them.

Q: How do I enable the API? A: Add --api to COMMANDLINE_ARGS; it exposes endpoints on the same port (7860).

Q: Why is the first generation slow? A: Models and optimizers are loaded/compiled on first use; subsequent generations are faster.

Q: Can I run it without a GPU? A: Yes with CPU, but it is extremely slow; only small SD1.5 models are practical.

Q: How do I update? A: Run git pull in the webui folder, then restart the launcher.

Q: What does --medvram do versus --lowvram? A: Both reduce VRAM use by offloading; --medvram is milder and faster, --lowvram is stricter for very limited GPUs.

Related Guides

• ComfyUI Fix & Optimization Guide
• Ollama Fix & Optimization Guide
• Dev RAM Calculator

Accuracy Note

Commands and paths reflect common, real-world setups as of 2026-08. Always verify against your installed version and OS. When in doubt, consult the official AUTOMATIC1111 documentation.