🎨 Image Generator with FLUX Stable Diffusion

This repository contains an image generator based on the FLUX model of Stable Diffusion, focused on providing high-quality image generation with various resource configurations.

📋 Requirements

• 🐍 Python 3.8 or higher
• 🔥 PyTorch 2.6.0 or compatible
• 🖥️ CUDA (recommended for GPU acceleration)
• 🤗 Hugging Face account (to obtain an access token)

💻 Installation

1. Clone this repository:

git clone https://github.com/R4F405/python-stable-diffusion-generator.git
cd python-stable-diffusion-generator

2. Create a virtual environment and install the dependencies:

python -m venv .venv
# On Windows
.venv\Scripts\activate
# On macOS/Linux
source .venv/bin/activate

pip install -r requirements.txt

⚙️ Configuration

Before running the generator, you need a Hugging Face access token:

1. Sign up at Hugging Face
2. Go to your profile → Settings → Access Tokens
3. Create a new token
4. Create a .env file in the project root with the following content:

   HUGGINGFACE_TOKEN=your_token_here
   

⚠️ Important: Never share your token or upload the .env file to public repositories.

🚀 Stable Diffusion Installation and Setup

This project uses the diffusers library, which provides a unified API for working with diffusion models like Stable Diffusion. The library automatically manages the download and installation of the required model.

🌊 FLUX Model

FLUX is an efficient implementation of Stable Diffusion that offers high-quality image generation with lower resource usage. The script specifically uses the "FLUX.1-schnell" model, which balances speed and quality.

🔄 First Run

The first time you run any of the scripts:

1. 📥 The FLUX model will be automatically downloaded from Hugging Face (approximately 2GB).
2. 💾 The model will be stored in the local Hugging Face cache (usually in ~/.cache/huggingface/).
3. ⏱️ This process may take a few minutes depending on your internet connection.

🖥️ Hardware Requirements

• With GPU (NVIDIA):

  • 🔋 Minimum: 4GB VRAM (using the script optimized with funcionales.py)
  • 🔋🔋 Recommended: 8GB+ VRAM for better performance
  • 🛠️ CUDA Toolkit installed (usually installed automatically with PyTorch if a compatible GPU is detected)

• Without GPU:

  • 💻 Minimum 16GB of system RAM
  • 🔄 CPU with good processing power
  • ⏱️ Be aware that generation will be significantly slower.

🔍 Usage

The repository contains two main scripts:

1. 🚀 GenerarImagenesFlux.py

Optimized script for general use that automatically detects if you have a GPU:

python GenerarImagenesFlux.py

2. 🧰 funcionales.py (Script with adjustments for limited VRAM)

This file contains a script optimized for systems with a GPU but limited VRAM (around 5GB):

python funcionales.py

🎛️ Customizable Parameters

You can modify the following parameters in the scripts to customize the generation:

• 💬 prompt: The descriptive text to generate the image.
• 🎯 guidance_scale: Controls how much the image adheres to the prompt (0.0 = more creative freedom).
• 🔄 num_inference_steps: Processing steps (more steps = better quality but more time).
• 📐 width and height: Dimensions of the generated image.
• 🎲 generator.manual_seed(): Seed for reproducibility of results.

⚡ Performance Tuning

• For systems with low VRAM:

  • 🔽 Reduce num_inference_steps to values between 4-20.
  • 🖼️ Use lower resolutions like 512x512.
  • 🧩 Enable enable_attention_slicing().
  • 📊 Use torch.float16 instead of torch.bfloat16.

• For systems without GPU:

  • 🖥️ The main script will automatically detect the absence of a GPU and use the CPU.
  • ⏱️ Be aware that CPU generation will be significantly slower.

🔧 Troubleshooting

• 🚫 CUDA out of memory error: Reduce the resolution, inference steps, or use the script optimized for limited VRAM.
• 🔌 Model errors: Ensure you have the correct Hugging Face token and a stable internet connection.
• 📉 Low-quality images: Increase the number of inference steps to improve quality.

📚 Available Models

The generator uses the "black-forest-labs/FLUX.1-schnell" model by default, but you can experiment with other Stable Diffusion models by modifying the line:

pipe = FluxPipeline.from_pretrained("black-forest-labs/FLUX.1-schnell", ...)

📄 License

This project is licensed under the MIT License. Consult the LICENSE file for more details.