How to Install Stable Diffusion: A Complete Setup Guide
Stable Diffusion is one of the most powerful open-source AI image generation tools available — and unlike cloud-based alternatives, it runs locally on your own hardware. That means no subscriptions, no usage limits, and full control over your outputs. But getting it running requires a few moving parts, and the right approach depends heavily on your system and comfort level with technical setup.
What Stable Diffusion Actually Is (Before You Install Anything)
Stable Diffusion is a latent diffusion model — an AI system trained to generate images from text prompts. Unlike tools that live entirely in a browser, the standard version of Stable Diffusion runs on your machine using your GPU or CPU to process image generation locally.
The core model itself is just a collection of files (primarily .safetensors or .ckpt format). To use it practically, you need a frontend interface — the most widely used being AUTOMATIC1111's Stable Diffusion Web UI, though alternatives like ComfyUI and InvokeAI are also popular.
This guide focuses on the most common installation path, with notes on where your own setup will shape your choices.
System Requirements: What You Actually Need
Installation success depends almost entirely on your hardware. Here's what matters:
| Component | Minimum (Functional) | Recommended |
|---|---|---|
| GPU VRAM | 4GB (limited output) | 8GB+ for comfortable use |
| RAM | 8GB | 16GB+ |
| Storage | 10GB free | 20GB+ (models are large) |
| GPU Type | NVIDIA (best support) | NVIDIA RTX series |
| OS | Windows 10/11, Linux, macOS | Windows or Linux |
NVIDIA GPUs have the strongest support because Stable Diffusion relies on CUDA, NVIDIA's parallel computing platform. AMD GPUs can work — particularly on Linux using ROCm — but setup is more involved and performance varies. Apple Silicon Macs (M1/M2/M3) can run Stable Diffusion via the MPS backend, though generation speed and feature availability differ from NVIDIA setups.
CPU-only generation is technically possible but extremely slow — we're talking minutes per image rather than seconds.
The Core Installation Path (AUTOMATIC1111 Web UI)
This is the most documented and community-supported route. Here's what the process looks like:
Step 1: Install Python
Stable Diffusion Web UI requires Python 3.10.x (specific version matters — 3.11+ can cause compatibility issues with some extensions). Download it from python.org and make sure to check "Add Python to PATH" during installation on Windows.
Step 2: Install Git
Git is used to download and update the Web UI. Install it from git-scm.com. This gives you the git command in your terminal.
Step 3: Clone the Repository
Open a terminal or command prompt and run:
git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui This downloads the Web UI codebase to your machine.
Step 4: Download a Model
The Web UI is just the interface — you need an actual model checkpoint to generate images. Models are distributed through sites like Hugging Face and Civitai. Place downloaded model files into the models/Stable-diffusion/ folder inside your Web UI directory.
Popular starting models include Stable Diffusion 1.5, SDXL (Stable Diffusion XL), and various community fine-tuned versions. Model file sizes typically range from 2GB to 7GB+.
Step 5: Launch the Web UI
On Windows, run webui-user.bat. On Linux/macOS, run webui.sh. The first launch will automatically download additional dependencies — this can take several minutes depending on your connection.
Once complete, the interface opens in your browser at http://127.0.0.1:7860.
🖥️ Alternative Frontends Worth Knowing
AUTOMATIC1111 isn't the only option. Your workflow preferences may point you elsewhere:
- ComfyUI — node-based interface, highly customizable, preferred by users who want granular control over the generation pipeline
- InvokeAI — cleaner UI, easier initial setup, good for users who want a more polished experience out of the box
- Fooocus — extremely simplified interface, designed to require minimal configuration
Each frontend installs differently and has varying extension/plugin ecosystems.
Where Things Get Variable 🔧
The installation process above is consistent in structure, but several factors will change your specific experience significantly:
GPU VRAM is the biggest constraint. Users with 4–6GB of VRAM will need to enable specific command-line arguments (--medvram or --lowvram) that reduce memory usage at the cost of speed. Users with 12GB+ can run larger models and generate at higher resolutions without adjustments.
Operating system changes the launch commands, Python installation steps, and which GPU acceleration backend is used. Linux users often have an easier time with driver configuration; Windows users benefit from more community troubleshooting resources.
Model choice affects both file size requirements and the type of imagery the system produces best. SDXL models produce higher-quality outputs but demand more VRAM and processing time than SD 1.5 models.
Technical comfort level matters too. Troubleshooting dependency errors, editing configuration files, or installing extensions requires comfort with the terminal. First-time users occasionally encounter errors during the initial dependency installation that require searching community forums like Reddit's r/StableDiffusion.
The Part That Depends on You
The installation process has a clear structure — Python, Git, the Web UI, a model, and a launch script. But which frontend suits your workflow, which model fits your creative goals, and how much hardware headroom you're working with will shape what "installed and working" actually looks like for you. Those answers sit with your specific machine and what you're trying to build with it.