zsxkib / sd3-on-apple-silicon

Run Stable Diffusion on Apple Silicon

Stable Diffusion 3 on Apple Silicon (MPS)

This code allows you to run Stable Diffusion 3 on Apple Silicon devices using the MPS backend.

Prerequisites

• Python 3.11
• Conda
• Hugging Face API token (optional)

Setup

1. Create a new Conda environment:

conda create -n sd3 python=3.11 -y
conda activate sd3

2. Install the required packages:

pip install -r requirements.txt

3. Set your Hugging Face API token (if required):

export HF_API_TOKEN=your_token_here

You can add this line to your .zshrc file to make it persistent.

Usage

1. Save the provided code as sd3-on-mps.py.

2. Run the script:

python sd3-on-mps.py

3. The generated image will be saved as sd3-output-mps.png in the same directory.

Notes

• The code automatically detects the available device (MPS, CUDA, or CPU) and uses the best one.
• If your system has less than 64 GB of RAM, the code enables attention slicing to reduce memory usage.
• The model cache is set to ./sd3-cache to avoid downloading the model repeatedly.
• The image generation process may take some time depending on your hardware.

Tips and Tricks

You can change how your images look in sd3-on-mps.py by adjusting these settings:

• seed: Set this to a specific number to make the same image again, or leave it as None to get different images each time.
• prompt: Change this text to create various images (for example, "A sunset over the mountains").
• height and width: Change these numbers to make the image bigger or smaller (for example, 1024 for bigger images).
• num_inference_steps: Use more steps for clearer images. More steps mean more detail but take more time.
• guidance_scale: Adjust this to influence how much the image matches the prompt. Higher values make the image more accurate but might look less natural.

Playing with these options lets you make many different kinds of images.

That's it! You can now generate images using Stable Diffusion 3 on your Apple Silicon device.