01/08/2025 - App runs, inference is much faster and only 20 steps produces fantastic results. Still optimising

11/06/2025 - App runs, inference is very slow. Busy fixing

LumaFlow-Lumina-WebUI - a highly optimised Lumina Image2.0 gradio web app for low VRAM systems (6GB+)

Welcome! This little app lets you harness the power of Lumina Image 2.0 for your text-to-image creations, right on your own computer. It's designed to be super easy to get started, especially if you're not a fan of complicated setups.

Think of it as your local, friendly interface to a rather smart image model, designed to be (relatively) kind to your GPU.

What's the Big Idea?

• Super Simple Setup: Just download, click install.bat, then launch.bat. That's the dream!
• Lumina Image 2.0 Power: Uses the impressive Alpha-VLLM/Lumina-Image-2.0 model.
• Local & Private: Your prompts and images stay on your machine.
• User-Friendly UI: Clicky buttons and sliders, not scary code.
• VRAM Aware: Includes optimizations like CPU offloading to help run on a wider range of GPUs.

Features

• Generate images from your text prompts.
• Adjust image dimensions (Lumina likes multiples of 64, around 1024x1024 total pixels is a good start).
• Control generation steps and guidance scale (CFG).
• Use specific seeds for consistency or let randomness surprise you.
• Images are saved automatically to an output folder.
• Sleek, modern interface.

Requirements (The Essentials)

Hardware:

• A dedicated NVIDIA GPU is strongly recommended (ideally with 6GB+ VRAM, though Lumina is a bit more forgiving than some giants).
• CPU Mode? Possible....but don't bother.

Software:

• Python (version 3.9+ recommended). Get it from python.org (tick "Add Python to PATH" on Windows).
• Git (optional, for cloning). Get it from git-scm.com.

The "Click-Click-Done" Installation (Windows)

This is the easy route for Windows users!

1. Download the App:

   • Go to the GitHub repository page: https://github.com/Raxephion/LumaFlow-Lumina-WebUI
   • Click the green "<> Code" button, then "Download ZIP".
   • Extract the downloaded ZIP file to a folder on your computer (e.g., C:\LuminaApp).

2. Run the Installer:

   • Navigate into the folder where you extracted the files.
   • Double-click install.bat.
   • A black window will appear and show progress. It will download Python libraries and set up a virtual environment. This might take a few minutes, especially the first time. Grab a beverage.
   • Wait until it says "Installation complete" and prompts you to "Press any key to continue . . .".

3. Launch the App:

   • After the installation is done and you've pressed a key to close the installer window, double-click launch.bat in the same folder.
   • The app will start. The first time, it will download the Lumina Image 2.0 model files (a few gigabytes). This is a one-time download per model.
   • Once ready, it will show a URL like http://127.0.0.1:7860. Open this in your web browser.

That's it! You should be ready to generate images.

Manual Installation (For Other OS or if .bat files aren't your jam)

If you're on macOS/Linux, or prefer doing things step-by-step:

1. Get the Code:

   • With Git:

     git clone https://github.com/Raxephion/LumaFlow-Lumina-WebUI.git
     cd LumaFlow-Lumina-WebUI

   • Download ZIP: As described in Step 1 of the "Click-Click-Done" section, then navigate into the extracted folder with your terminal.

2. Create a Virtual Environment: In your terminal, inside the project folder:

   python -m venv venv

3. Activate the Virtual Environment:

   • Windows (in cmd or PowerShell): venv\Scripts\activate
   • macOS/Linux (in bash/zsh): source venv/bin/activate (Your terminal prompt should change to show (venv))

4. Install Dependencies: With the virtual environment active:

   pip install -r requirements.txt

5. Run the App:

   python app.py

   Look for the http://127.0.0.1:7860 URL in the terminal output and open it in your browser.

About Lumina Image 2.0

This app uses Alpha-VLLM/Lumina-Image-2.0. It's part of the Lumina family of models, which are "next-generation foundation models for text-to-image generation, text-to-video generation, and multi-modal language understanding." You can find more details on their Hugging Face Hub page.

Troubleshooting

• "CUDA out of memory": Try smaller image dimensions. Close other GPU-hungry apps.
• Slow first launch: Probably downloading the model. Be patient.
• Errors during install/launch:
  • Ensure Python is installed correctly and added to PATH.
  • Make sure you're running commands inside the activated virtual environment for manual setup.
  • If install.bat fails, check its output for specific error messages.

Disclaimer

This is a user-friendly wrapper for a powerful model. Use it responsibly. Image quality and performance can vary. Have fun experimenting!

Thanks

Mad respect to: 🧠 QIN QI (ChinChyi) and stzhao (zhaoshitian) from Alpha-VLLM — your work is seriously inspiring and foundational to projects like this. You bring the sorcery, I bring the hype :)