Project Overview

using https://arxiv.org/abs/2309.05019 SA-Solver Scheduler

This project demonstrates how to generate images using Stable Diffusion XL with optional prompt enhancement via Gemini's API. The code includes configurable settings and the ability to save/load these settings from a JSON file.

Prerequisites

Python 3.8 or higher (Note: For using torch.compile, Python 3.11 or lower is required)
venv or conda for creating a virtual environment

Setup

Using `venv`

Create a virtual environment:
```
python -m venv civitai_env
```

Activate the virtual environment:

On Windows:
```
civitai_env\Scripts\activate
```

On macOS and Linux:

source civitai_env/bin/activate

 source civitai_env/bin/activate.fish

Install the required libraries:

pip install torch diffusers accelerate pillow google-generativeai transformers

Using `conda`

Create a conda environment:

conda create --name civitai_env python=3.11

Activate the conda environment:

conda activate civitai_env

Install the required libraries:

conda install pytorch torchvision torchaudio pytorch-cuda=11.7 -c pytorch -c nvidia

pip install diffusers accelerate pillow google-generativeai transformers

Usage

Clone the repository:

git clone https://github.com/al-swaiti/SDXL-LOCAL

cd SDXL-LOCAL

Run the script:

python main.py

Configuration

The script uses a configuration file (config.json) "auto created after first run"to store user inputs. If the file doesn't exist, default values are used. The script will prompt you to enter the following details:

Model location
Prompt
Negative prompt
Width and height of the image
Number of inference steps
CFG scale
Seed
Gemini API key (optional)

You can also provide these inputs through the command line. If no input is provided, the default values or the values from the configuration file will be used.

Enhancing Prompts with Gemini's API

To get professional prompt results, you can use the Gemini API. If you wish to use this feature, insert your Gemini API key when prompted or include it in the configuration file.

Notes

If you want to use torch.compile, ensure you are using Python 3.11 or lower. and enable ( pipe.unet = torch.compile(pipe.unet, mode="reduce-overhead", fullgraph=True) ) line.
The generated image will be saved with a timestamp-based filename.
for low gpu disable ( pipe.to("cuda")) and enable (pipe.enable_sequential_cpu_offload())

follow me on https://civitai.com/user/AbdallahAlswa80

watch me on https://www.deviantart.com/abdallahalswaiti

connect me on https://www.linkedin.com/in/abdallah-issac/

try my model https://civitai.com/models/130664