How to Install Stable Diffusion Locally: A UK Guide for 2026

Introduction

Stable Diffusion has become the go-to toolkit for AI image generation. In 2026, the technology has matured significantly, but installing it locally remains a rite of passage for enthusiasts and professionals alike. Why bother with a local install when there are so many cloud services? Control, privacy, cost, and the ability to fine-tune models without restrictions are powerful motivators. This guide will walk you through installing Stable Diffusion on your own machine, using practical steps and clear examples.

Whether you're a digital artist, a developer, or just a curious tinkerer in the UK, you'll find everything you need to get started. We'll cover two of the most popular interfaces: Automatic1111 Web UI and ComfyUI. Both are free, open-source, and actively maintained. Let's dive in.

What You'll Need

Before we start, let's talk about hardware. Stable Diffusion is compute-hungry, but you don't need a datacenter. In 2026, the minimum viable setup is:

• GPU: Nvidia RTX 2060 (6GB VRAM) or better. AMD cards work with ROCm on Linux, but for Windows, Nvidia remains the smoothest path. Apple Silicon (M1/M2/M3/M4) with 16GB unified memory is also viable.
• RAM: 16GB system RAM is the sweet spot; 8GB will be painful.
• Storage: At least 20GB of free space. Models (checkpoints) typically range from 2GB to 7GB each.
• OS: Windows 10/11, Linux, or macOS. We'll cover Windows and Linux/macOS in one go.

Software-wise, you need Python 3.10/3.11 and Git. In 2026, Python 3.12 is supported by most projects, but 3.11 remains safest for compatibility. We'll also use a package manager like pip.

Important for UK users: If you're on a metered connection (e.g., mobile broadband), be aware that downloading models can be large. Also, check your electricity costs—Stable Diffusion can draw significant power. Consider using a free tier of a cloud GPU if your hardware doesn't cut it.

Method 1: Installing Automatic1111 Web UI (Easiest)

Automatic1111 is the most popular web interface. It's feature-rich and has a huge community. Here's how to install it locally.

Step 1: Install Prerequisites

First, install Python and Git if you don't have them.

• Windows: Download Python from python.org and tick *Add Python to PATH*. For Git, use git-scm.com and accept the defaults.
• Linux (Debian/Ubuntu): Use sudo apt install python3.10 python3.10-venv git.
• macOS: Use Homebrew: brew install [email protected] git.

Step 2: Clone the Repository

Open a terminal (Command Prompt on Windows, Terminal on Linux/macOS) and navigate to where you want the installation. Run:

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

Step 3: Run the Setup Script

The repository includes a webui.sh (Linux/mac) or webui-user.bat (Windows) script. On Windows, double-click webui-user.bat; on Linux/mac, run ./webui.sh. The first run will create a virtual environment, install dependencies, and download the default model (which can take a few minutes).

Troubleshooting tip: If the default model download fails (it often does due to network issues), you can manually download a model from Hugging Face or Civitai and place it in models/Stable-diffusion/.

Step 4: Access the Web UI

Once the script finishes, you'll see a local URL like http://127.0.0.1:7860. Open that in your browser. You're now running Stable Diffusion!

Basic Usage Example

At the top, you have a prompt box. Try something like a photograph of a red phone box in London, rainy, cinematic lighting. Negative prompt: blurry, low quality, watermark. Set steps to 20, sampler to Uniform, and CFG scale to 7. Click Generate. In a few seconds, you'll have an image.

Managing Models and Extensions

• To add new checkpoints, place them in models/Stable-diffusion/. They appear in the dropdown at the top-left after refreshing.
• To install extensions, go to the *Extensions* tab, select *Install from URL*, and paste a GitHub URL. Popular extensions include ControlNet, LoRA, and upscalers.

Method 2: ComfyUI (For Power Users)

ComfyUI is a node-based interface that gives you granular control. It runs faster but has a steeper learning curve. As of 2026, ComfyUI is popular among professionals for its efficiency and flexibility.

Installation

1. Clone the repository:
2. ```bash
3. git clone https://github.com/comfyanonymous/ComfyUI.git
4. cd ComfyUI
5. ```
6. Install dependencies:
7. ```bash
8. pip install -r requirements.txt
9. ```
10. Or on Windows, you can use the provided installer folder.
11. Run it:
12. - Windows: python main.py
13. - Linux/macOS: python main.py --dont-print-server-header
14. Open http://127.0.0.1:8188.

First Workflow Example

ComfyUI opens with a default workflow: it loads a checkpoint and generates an image from a prompt. You'll see nodes for Load Checkpoint, CLIP Text Encode, KSampler, VAE Decode, and Save Image. Connect them if they're not pre-connected. Set your prompt in the text node, pick a checkpoint, and click Queue Prompt. The graph will execute, and the image will pop out.

Why Choose ComfyUI?

• Speed: It uses less VRAM and processes faster.
• Reproducibility: You can save workflows as JSON files and share them.
• Advanced control: You can build complex pipelines for video, inpainting, and multi-model ensembling.

But if you're a beginner, start with Automatic1111. It's more forgiving.

Hardware Optimisation and Performance

In 2026, Stable Diffusion supports the latest GPUs, but you can also use consumer cards effectively. Here are some tips to get the most out of your hardware:

Enable xformers or SDP yes In Automatic1111, add --xformers or --opt-sdp to the command line arguments in webui-user.bat (Windows) or webui.sh (Linux). This speeds up attention computations and reduces VRAM usage.

Use --medvram or --lowvram If you have 4GB of VRAM or less, use --medvram or --lowvram. These flags trade speed for memory by splitting the model across CPU and GPU.

Upscaling and Tiling For high-resolution images, use the Hires. Fix option in Automatic1111, or add an upscale workflow in ComfyUI. This generates a base image and then upscales it in chunks.

Ampere and Blackwell Optimisations If you have a newer Nvidia card (RTX 30/40/50 series), you can use TensorRT extensions for massive speed boosts. ComfyUI has a dedicated TensorRT node pack, and Automatic1111 has a stable TensorRT extension.

Choosing the Right Model

The default model (*Stable Diffusion 2.1*) is fine, but in 2026, you have better options. Popular checkpoints include:

• SDXL 1.0 / SDXL Turbo: Great for high-quality and fast generation. Requires ~8GB VRAM, but you can run with --medvram.
• Realistic Vision: Superb for photorealistic people and scenes.
• Proteus / DreamShaper: Versatile for illustration and fantasy.

Where to find models?

• Hugging Face: Official model hub with thousands of checkpoints.
• Civitai: A huge community for fine-tuned models and LoRAs. Always check the license.

UK-specific tip: Use the --listen flag to access your instance from a tablet or secondary computer over your local network. Just be careful about exposing it to the internet—you'll want to set authentication or use a VPN.

Common Pitfalls and Fixes

"CUDA Out of Memory" Reduce batch size, use lower resolution, or enable medvram. Also ensure no other GPU-intensive apps are running.

"Could not find the PyTorch package" Install PyTorch with the correct version. For CUDA, use the command from pytorch.org (e.g., pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121).

Slow Generation on Apple Silicon Use the --device mps if using Automatic1111 on macOS. For ComfyUI, it's automatic. You may also want to use a --opt-sdp for reduced memory.

Antivirus Interference Windows Defender sometimes flags xformers or GIT files. Add an exclusion for your installation folder.

Automation and Prompting

Beyond the GUI, you can script Stable Diffusion for batch processing. Both Automatic1111 and ComfyUI have APIs. With a few lines of Python, you can generate thousands of images automatically.

Example using Automatic1111's API:

```python import requests import json

url = "http://127.0.0.1:7860/sdapi/v1/txt2img" payload = { "prompt": "a British cottage garden, lavender, soft sunlight", "negative_prompt": "dark, blurry, oversaturated", "steps": 25, "width": 512, "height": 512, "sampler_name": "UniPC", "cfg_scale": 7.5 } response = requests.post(url, json=payload) with open("output.png", "wb") as f: f.write(base64.b64decode(response.json()["images"][0])) ```

This opens the door to automatic art generation, content creation, and even quick mockups for presentations.

Staying Up-to-Date

The AI ecosystem moves fast in 2026. To keep your installation current:

• In Automatic1111, click Update in the footer.
• For ComfyUI, run git pull and then pip install -r requirements.txt again.

Watch out for breaking changes. Before updating, check the project's changelog and back up your models and embeddings folders.

Alternative: Using a Cloud GPU

If your hardware can't handle it, or you want to test before committing, consider using a cloud service. In the UK, popular options include:

• Google Colab: Free tier with a T4 GPU. Enough for casual use.
• RunPod: Commercial, but you can use prebuilt Stable Diffusion images.
• AWS EC2 G4dn: A bit complex but flexible.

They're not local, but they follow the same install processes, so you'll be ready when you upgrade your PC.

Conclusion

Installing Stable Diffusion locally is a skill that pays off. You get unlimited creative freedom, no restrictions on content (within legal bounds), and the ability to experiment with cutting-edge models. In this guide, you've learned two methods: the easy path with Automatic1111, and the flexible path with ComfyUI. You now know how to choose models, optimise performance, and troubleshoot common issues.

The best way to learn is by doing. Start with a simple prompt, generate something, and then tweak. Try adding a negative prompt, adjust the steps, or install a ControlNet extension. The community around Stable Diffusion is vast, and there are countless tutorials and resources available.

Remember, the future of AI is in your hands—or rather, on your own GPU. Happy generating!

FAQ

1. Do I need a high-end GPU to run Stable Diffusion locally?

No, but you'll need at least an Nvidia GPU with 6GB VRAM for a decent experience. You can run with 4GB using --medvram, but generation will be slower. Apple Silicon with 16GB unified memory works well too.

2. How much disk space do I need?

At least 20GB. The base installation plus a few models will quickly consume 15-20GB. If you plan to download many checkpoints, consider a dedicated SSD.

3. Is it legal to use Stable Diffusion in the UK?

Yes, Stable Diffusion is open-source and legal to use. However, you must respect the licenses of individual models. Also, be aware of copyright laws when generating images of real people or trademarked characters—commercial use may have restrictions.

4. Can I run Stable Diffusion on a laptop?

Yes, if it has a decent GPU or Apple Silicon. Expect higher temperatures and shorter battery life. For gaming laptops with RTX 3060 or better, performance is great. For ultrabooks with integrated graphics, it might be too slow.