fluxflow-ui 0.8.0

Web UI for FluxFlow text-to-image generation

FluxFlow UI

Web interface for FluxFlow text-to-image generation and training.

🚧 Model Availability Notice

Training In Progress: FluxFlow models are currently being trained. The UI is fully functional, but trained model checkpoints are not yet available for download.

When Available: Trained checkpoints will be published to MODEL_ZOO.md upon completion of training validation.

Current Capabilities: You can use this UI to:

• Configure and launch training runs with your own datasets
• Monitor training progress in real-time
• Test the architecture with your own trained checkpoints

---

Installation

Note: This documentation describes v0.8.0. For the previous stable version (v0.4.0), see v0.4.0 documentation.

Prerequisites

Required:

• Python 3.10 or higher
• pip package manager
• 8GB+ RAM (16GB+ recommended)

For GPU Training (Recommended):

• NVIDIA GPU: CUDA 11.8+ with compatible drivers
• Apple Silicon: macOS 12.3+ (for MPS support)
• GPU VRAM: 8GB minimum, 16GB+ for high quality training

Verify Prerequisites:

python --version  # Should show 3.10 or higher
pip --version     # Should be installed
nvidia-smi        # (NVIDIA only) Should show GPU info

Production Install (v0.8.0 - Current Stable)

pip install fluxflow-ui

What gets installed:

• fluxflow-ui - Web interface for training and generation
• fluxflow-training - Training capabilities (automatically installed as dependency)
• fluxflow core package (transitively installed)
• CLI command: fluxflow-ui

Package available on PyPI: fluxflow-ui v0.8.0

⚠️ Note: v0.8.0 requires fluxflow-training>=0.8.0 and supports the pillar-attention flow architecture introduced in v0.8.0.

Development Install

git clone https://github.com/danny-mio/fluxflow-ui.git
cd fluxflow-ui
pip install -e ".[dev]"

⚠️ Security Warning

FluxFlow UI is designed for local development use only.

• No authentication or authorization
• File browser can access entire filesystem
• Not hardened for production deployment

See SECURITY.md for details on security measures, limitations, and production deployment warnings.

Do not expose this application to the internet without additional security hardening.

---

Features

• Training Interface: Configure and monitor training runs
  • Start/stop training, monitor loss curves and metrics, view sample generations
• Generation Interface: Generate images from text prompts
  • Load trained models, adjust sampling parameters, batch generation support
• Real-time Progress: Live training updates and logs
• Model Management: Load and manage checkpoints
• Interactive Controls: Adjust generation parameters in real-time

Quick Start

Launch the Web UI

FluxFlow UI supports two interfaces:

Flask (Primary - Recommended):

fluxflow-ui

Gradio (Alternative):

python -m fluxflow_ui.app

Then open your browser to http://localhost:7860

Note: Flask is the primary interface with full features. Gradio is provided as an alternative but may have limited functionality.

Classifier-Free Guidance (CFG)

✨ Available since v0.3.0: FluxFlow UI supports training and generation with Classifier-Free Guidance.

Training with CFG

To train models with CFG support:

1. Navigate to the Training tab
2. Expand the CFG Training section
3. Set cfg_dropout_prob between 0.0-0.20 (recommended: 0.10-0.15)
   • This randomly drops text conditioning during training
   • Higher values = stronger CFG effect but may reduce unconditional quality
   • Set to 0.0 to disable CFG training

Generating with CFG

To use CFG during generation:

1. Navigate to the Generation tab
2. Load a checkpoint trained with cfg_dropout_prob > 0
3. Expand the CFG Settings section
4. Enable CFG and set parameters:
   • Enable CFG: Toggle on
   • Guidance Scale: 1.0-15.0 (recommended: 3.0-7.0)
     • 1.0 = no guidance
     • 3.0-7.0 = balanced quality/creativity
     • 7.0-15.0 = strong guidance (may oversaturate)
   • Negative Prompt (optional): Text to avoid in generation

Note: CFG requires 2× forward passes per sampling step, doubling generation time.

CFG Benefits

• Better prompt adherence: Images follow text descriptions more closely
• Higher quality: Improved coherence and detail
• Negative prompts: Ability to steer away from unwanted features
• Flexible control: Adjust guidance strength per generation

Package Contents

• fluxflow_ui.tabs - UI tab implementations
• fluxflow_ui.utils - Config management and training runners
• fluxflow_ui.templates - HTML templates
• fluxflow_ui.static - CSS and JavaScript assets

Configuration

The UI runs on http://0.0.0.0:7860 by default. To customize the host and port, modify the main() function in src/fluxflow_ui/app_flask.py.

Development

Install with development dependencies:

pip install -e ".[dev]"

Links

• GitHub Repository
• Security Policy
• User Guide
• Troubleshooting

License

MIT License - see LICENSE file for details.