How to Intermediate · 3 min read

How to run Stable Diffusion locally

Q: How to run Stable Diffusion locally

To run Stable Diffusion locally, install Python 3.8+, set up a virtual environment, and install the diffusers library from Hugging Face. Use the pre-trained model weights to generate images offline without internet access.

Quick answer

To run Stable Diffusion locally, install Python 3.8+, set up a virtual environment, and install the diffusers library from Hugging Face. Use the pre-trained model weights to generate images offline without internet access.

PREREQUISITES

Python 3.8+
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117
pip install diffusers transformers scipy
A GPU with at least 6GB VRAM (recommended)
Git (optional for cloning repos)

Setup

Install Python 3.8 or higher and create a virtual environment. Then install PyTorch with CUDA support (if you have an NVIDIA GPU) and the Hugging Face diffusers library for Stable Diffusion.

bash

python -m venv sd-env
source sd-env/bin/activate  # Linux/macOS
sd-env\Scripts\activate  # Windows
pip install --upgrade pip
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117
pip install diffusers transformers scipy

Step by step

Use the following Python script to load the Stable Diffusion model locally and generate an image from a text prompt.

python

import os
from diffusers import StableDiffusionPipeline
import torch

# Load model locally with caching
model_id = "runwayml/stable-diffusion-v1-5"
pipeline = StableDiffusionPipeline.from_pretrained(model_id, torch_dtype=torch.float16)
pipeline = pipeline.to("cuda")

prompt = "A fantasy landscape, vivid colors"
image = pipeline(prompt).images[0]

image.save("output.png")
print("Image saved as output.png")

output

Image saved as output.png

Common variations

Use CPU only by changing pipeline.to("cuda") to pipeline.to("cpu"), but expect slower generation.
Try different models by changing model_id to other Hugging Face Stable Diffusion variants.
Use torch.autocast for mixed precision to optimize VRAM usage.

python

import torch

with torch.autocast("cuda"):
    image = pipeline(prompt).images[0]

Troubleshooting

If you get CUDA out of memory errors, reduce batch size or switch to CPU mode.
Ensure your GPU drivers and CUDA toolkit are up to date.
If model download fails, check your internet connection or manually download weights from Hugging Face.
For permission errors, run your terminal as administrator or check file system permissions.

✅

Key Takeaways

Use Python 3.8+ and install PyTorch with CUDA for best performance running Stable Diffusion locally.
The Hugging Face diffusers library provides easy access to pre-trained Stable Diffusion models for offline use.
GPU with at least 6GB VRAM is recommended for smooth image generation; CPU mode is slower but possible.
Adjust precision and device settings to optimize memory and speed based on your hardware.
Keep your GPU drivers and CUDA toolkit updated to avoid runtime errors.

Verified 2026-04 · runwayml/stable-diffusion-v1-5

Verify ↗