How to Install Stable Diffusion Locally (2026 Guide)

Learning how to install Stable Diffusion locally gives you something no cloud service can: unlimited, private, free AI image generation on your own hardware. No subscriptions, no content filters beyond your own judgment, no queue times, and complete privacy for your prompts and images. The setup takes some effort — but it’s very doable, even if you’re not deeply technical. Here’s the full guide.

Why Run Stable Diffusion Locally?

• Cost: after the hardware, generation is free and unlimited. Heavy users save enormously versus subscriptions.
• Privacy: prompts and images never leave your machine. Essential for sensitive, proprietary, or personal work.
• Control: access to custom models, LoRAs, and settings cloud services don’t offer. Our explainer on what LoRAs are covers the customization rabbit hole.
• Offline capability: generate on planes, in remote locations, or anywhere without reliable internet. Our guide to local AI image generation explores the broader offline setup.
• Learning: running the model yourself teaches you how diffusion actually works in a way cloud interfaces never will.

Hardware Requirements: The Honest Version

The GPU is everything. Stable Diffusion runs on your graphics card’s VRAM:

• Minimum viable: 6GB VRAM (older cards like GTX 1660). Works, but slow and limited to smaller images.
• Comfortable: 8GB VRAM (RTX 3060 Ti, 4060 Ti). Good speed at standard resolutions, handles most models.
• Ideal: 12GB+ VRAM (RTX 4070 and up). Fast, handles SDXL and larger models comfortably. Our guide to VRAM for Stable Diffusion XL details the requirements for the bigger models.
• CPU/RAM: 16GB system RAM recommended; CPU matters less since the GPU does the work.
• Storage: models are 2–7GB each, and you’ll accumulate them. Budget 50GB+ if you catch the collecting bug.
• Mac users: Apple Silicon Macs (M1 and later) run Stable Diffusion via MPS acceleration — slower than equivalent NVIDIA GPUs but perfectly usable. Our article on MacBooks and local LLMs touches on the unified memory advantage.

NVIDIA GPUs work best due to CUDA support. AMD is possible but fussier; Intel Arc is emerging. When in doubt, NVIDIA is the path of least resistance.

How to Install Stable Diffusion Locally: The Easy Path (AUTOMATIC1111)

The most popular interface is Stable Diffusion WebUI (AUTOMATIC1111) — a local web app that gives you point-and-click generation with enormous depth when you want it.

On Windows

1. Install Python 3.10 (from python.org — check “Add to PATH” during installation) and Git.
2. Open a terminal and clone the WebUI: git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git
3. Download a model checkpoint (e.g., from Hugging Face or Civitai) and place the .safetensors file in models/Stable-diffusion/.
4. Run webui-user.bat. First launch downloads dependencies (takes a while — be patient).
5. Open http://127.0.0.1:7860 in your browser. You’re generating.

On Mac

1. Install Homebrew, then brew install python@3.10 git.
2. Clone the WebUI repository as above.
3. Add your model file to models/Stable-diffusion/.
4. Run ./webui.sh and open the local URL it provides.
5. Apple Silicon users: the script handles MPS configuration automatically in recent versions.

On Linux

Similar to Mac: install Python 3.10 and Git via your package manager, clone, add model, run ./webui.sh. Linux with NVIDIA drivers is arguably the smoothest experience.

The Even Easier Path: One-Click Installers

If the manual route sounds like too much, several polished installers bundle everything:

• Stability Matrix: manages multiple AI apps (not just Stable Diffusion) with one-click installs and updates. Great for explorers.
• ComfyUI portable builds: for those wanting node-based workflows without setup friction.
• Draw Things (Mac/iOS): a native Apple app that runs Stable Diffusion with zero terminal work.

These sacrifice some flexibility for convenience — a fair trade when starting out. You can always graduate to manual installs later.

Essential First Steps After Installation

1. Generate your first image. Default settings, simple prompt. Confirm everything works before tweaking.
2. Learn the key settings: sampling steps (20–30 is the sweet spot), CFG scale (7 is default; lower = more creative, higher = more literal), and sampler choice (DPM++ 2M Karras is a reliable default).
3. Try a second model. Different checkpoints have wildly different styles. Download one photorealistic and one artistic model to feel the range.
4. Install 2–3 essential extensions via the built-in extensions tab — but don’t go overboard initially. Learn the base first.
5. Set up a sensible file structure for outputs before you generate your thousandth image and can’t find anything.

Troubleshooting Common Issues

• “Out of memory” errors: lower the resolution, enable memory optimizations (--medvram or --lowvram flags), or close other GPU-using apps.
• Slow generation: check that you’re actually using GPU (not CPU fallback), update GPU drivers, try the --xformers flag for significant speedups on NVIDIA.
• Black images: usually a precision issue — try --no-half or updating PyTorch.
• Model won’t load: verify the file isn’t corrupted (check file size against the source), ensure it’s in the right folder, and confirm it’s a compatible format.

The Takeaway

Installing Stable Diffusion locally is a weekend project with permanent payoff: free, private, unlimited image generation with deep customization. Check your VRAM, pick the easy path (WebUI or a one-click installer), generate your first image, then explore models and settings at your pace. The initial setup is the hardest part — everything after is creative play. And once you’re comfortable, the customization world of LoRAs, inpainting, and advanced workflows opens up entirely.