How to Run Stable Diffusion Locally: A Step-by-Step Setup

You've halfway through a project, you'd like to mock up a quick visual, and uploading the client's reference photo to a public AI service feels risky. There's a straightforward fix: install Stable Diffusion on your own computer. Everything — your prompt, your reference image, the output picture — stays on your hard drive. No cloud, no uploads, no surprise terms-of-service changes.

This guide walks you through the install, the first image, and a basic refinement, with exact clicks and menu names. There's one honest caveat at the end — older or low-spec computers will struggle, and I'll tell you how to check before you start.

Before you start: what you'll need

Name every requirement up front so nothing surprises you mid-install:

• A computer from roughly the last 4–5 years. Stable Diffusion runs on your graphics card (the chip that normally powers your screen and games). A dedicated NVIDIA graphics card with at least 4 GB of VRAM (the graphics card's working memory) is the smoothest path. AMD cards work with extra setup. Macs with Apple Silicon (M1, M2, M3, M4 chips) work through a separate path I won't cover in detail here.
• About 10 GB of free disk space for the tool plus one starter model.
• Python 3.10 installed on your computer (a programming language many tools are built on — you'll install it once and never touch it again).
• Git installed (a tool for downloading code from the internet — also install once).
• A patient hour. The first install is slower than the hundredth.

If you don't have a dedicated NVIDIA graphics card, the tool still runs on your regular processor, just much more slowly — minutes per image instead of seconds. The steps below assume an NVIDIA card.

---

Step 1 — Install Python and Git

Open your web browser and go to the official Python download page at python.org. Download Python 3.10.x (a specific version number — newer 3.11+ versions can cause errors with this tool). Run the installer. Important: on the first installer screen, tick the checkbox that says "Add Python 3.10 to PATH" before you click Install Now. Without that tick, the next steps won't recognize Python.

Next, go to git-scm.com and download Git for your operating system. Run the installer with the default settings — you don't need to change anything.

💬 Example path you'll see: C:\Users\YourName\AppData\Local\Programs\Python\Python310\

You'll know it worked when you open a terminal window (Command Prompt on Windows, Terminal on Mac/Linux), type python --version, and the reply starts with Python 3.10.

---

Step 2 — Download Stable Diffusion WebUI

Open your terminal window. Type this exact command and press Enter:

💬 git clone https://github.com/AUTOMATIC1111/stable-diffusion-webui.git

This creates a new folder called stable-diffusion-webui in your current directory. Inside it, on Windows, you'll find a file called webui-user.bat. On Mac or Linux, the equivalent file is webui.sh. That's the file you'll double-click every time you want to start the tool.

You'll know it worked when the folder appears and contains files like webui.py, modules/, and models/.

---

Step 3 — Add your first model file

The tool is now a shell. It needs a model — the trained AI "brain" — to actually generate pictures. A solid starter model is Stable Diffusion 1.5 (sometimes labeled "SD 1.5") from the official Hugging Face page. The file is large (about 4 GB) and ends in .safetensors.

💬 Menu path you'll see: stable-diffusion-webui → models → Stable-diffusion/

Download the .safetensors file and drop it into that Stable-diffusion folder. Don't unzip it — leave the file as-is.

You'll know it worked when you see the .safetensors file sitting next to any other files in the Stable-diffusion folder, and when the file size is roughly 4 GB.

---

Step 4 — Start the tool and open the interface

Double-click webui-user.bat (Windows) or run ./webui.sh (Mac/Linux). A black terminal window will pop up and start downloading a few extra pieces. The first launch can take 5–15 minutes. Watch for the line that says something like Running on local URL: http://127.0.0.1:7860.

Open your web browser and go to that address. The Stable Diffusion WebUI loads — a simple gray page with a text box at the top labeled "Prompt", a second text box labeled "Negative prompt", and a "Generate" button on the left.

You'll know it worked when you see the prompt text box and a working Generate button. If you see a blank page or an error about missing modules, close the terminal, reopen it, and try again — the second launch usually finishes the setup.

---

Step 5 — Write your first prompt and generate

In the top text box (Prompt), describe the picture you want. In the second text box (Negative prompt), list what you don't want. Keep the prompt simple at first.

💬 Example prompt: a watercolor illustration of a cat sleeping on a stack of books, soft morning light

💬 Example negative prompt: blurry, distorted, extra limbs, text, watermark

Leave the other settings on their defaults for now. Click Generate. The image appears on the right side of the page, usually within 5–30 seconds depending on your hardware. The picture is saved automatically to stable-diffusion-webui/outputs/txt2img-images/ — find it through your file explorer.

You'll know it worked when an image preview appears on the right side of the page, and a .png file shows up in the txt2img-images folder.

---

Step 6 — Refine the image with img2img

The first image is rarely perfect. Scroll down below the Generate button until you see a panel labeled "img2img". Click the img2img tab.

Upload the image you just saved (drag it onto the upload area, or click the image icon). In the prompt box, describe what you want to change — for example, "warmer colors" or "sharper details". Set "Denoising strength" (a slider that controls how much the AI changes your image — 0.3 to 0.5 for gentle tweaks, 0.7 or higher for bigger changes). Click Generate.

💬 Example menu path: img2img tab → upload area → Denoising strength slider → Generate

You'll know it worked when the new image looks recognizably similar to the original but with your requested changes applied.

---

Common mistakes

• Forgetting to tick "Add Python to PATH" during install. The terminal can't find Python, and every later command fails. Fix: uninstall Python, reinstall, and tick the box this time.
• Launching the tool before the model file is in place. The interface opens but every generation returns an error about a missing model. Fix: confirm the .safetensors file is inside models/Stable-diffusion/, not just inside the main stable-diffusion-webui folder.
• Closing the black terminal window mid-generation. The tool shuts down and you lose your work. Fix: leave the terminal open as long as you're generating. Close it only when you're done for the session.
• Picking a model file from an untrusted source. Some websites repackage model files with extra code. Fix: download only from the official Hugging Face repository for AUTOMATIC1111's recommended models.
• Generating at very high resolution right away. Large images can run out of VRAM and produce a black image. Fix: start with the default 512×512 pixels, then upscale once you have a result you like.