Simple Diffusion Server

Introduction

simple-diffusion-server is a lightweight Flask-based server designed to generate images using the Stable Diffusion model. It offers a simple API endpoint to receive image generation requests and supports custom UNet models and LoRA adjustments.

Installation

Prerequisites

• Python 3.8 or higher
• Conda environment (recommended)

Setup

1. Clone the repository:

   git clone https://github.com/x0rb0t/simple-diffusion-server.git
   cd simple-diffusion-server

2. Create and activate a Conda environment (optional but recommended):

   conda create -n diffusion-server python=3.9
   conda activate diffusion-server

3. Install dependencies:

   pip install flask diffusers torch pillow

   If using Gunicorn for production:

   pip install gunicorn

4. Environment Variables: Optionally, set the environment variables for the default model, UNet model, LoRA directories, and scales:

   export MODEL_NAME="stabilityai/stable-diffusion-xl-base-1.0"
   export UNET_MODEL="latent-consistency/lcm-sdxl" 
   export LORA_DIRS="lora1.safetensors:lora2.safetensors"
   export LORA_SCALES="0.7:0.7"

Usage

Starting the Server

Run the server using Flask (for development):

python server.py

Or using Gunicorn (for production):

gunicorn -b 0.0.0.0:3101 server:app

Making Requests

To generate an image, send a POST request to the /generate-image endpoint with the desired parameters. Example using curl:

curl -X POST http://localhost:3101/generate-image \
     -H "Content-Type: application/json" \
     -d '{
         "prompt": "a surreal landscape, digital art",
         "num_inference_steps": 30,
         "guidance_scale": 8,
         "width": 1024,
         "height": 1024,
         "format": "jpeg"
         }'

Note: We are using latent-consistency/lcm-sdxl, so num_inference_steps is small

Command-Line Arguments

You can also start the server with specific model configurations using command-line arguments:

python server.py --model "your_model_name" --unet "your_unet_model" --lora_dirs "dir1:dir2" --lora_scales "0.7:0.8"

Additional Notes

• Ensure that the models and LoRA directories you specify are compatible with the diffusers library.
• For production deployment, consider using a reverse proxy like Nginx and setting up HTTPS.