ComfyUI Extension: ComfyUI Flux 1.1 Ultra & Raw Node

Authored by ShmuelRonen

Created

Updated

58 stars

A ComfyUI custom node for Black Forest Labs' FLUX 1.1 [pro] API, supporting both regular and Ultra modes with optional Raw mode.

Custom Nodes (1)

README

ComfyUI Flux Pro Integrative - Enhanced Flux API Node

Support My Work

If you find this project helpful, consider buying me a coffee:

Buy Me A Coffee

A completely rewritten and enhanced custom node for ComfyUI that integrates with Black Forest Labs FLUX API, providing seamless access to FLUX's image generation and finetuning capabilities with improved reliability and user experience.

BFL

✨ What's New in v2.0

Complete Rewrite with Modern Architecture:

  • 🏗️ Modular Design: Separated concerns with dedicated classes for configuration, API handling, and node logic
  • 🛡️ Robust Error Handling: Custom exception handling with detailed error messages and graceful fallbacks
  • 🔧 Smart Configuration: Automatically searches multiple locations for config files with clear setup instructions
  • 📝 Enhanced Logging: Detailed console output for better debugging and monitoring
  • 🎨 Better UX: Improved tooltips, organized parameters, and clearer status information
  • 🚀 Modern ComfyUI: Full compatibility with latest ComfyUI versions using modern node patterns

🌟 Features

Core Image Generation:

  • Support for FLUX 1.1 Pro regular and Ultra modes
  • Optional Raw mode for more natural-looking images
  • Multiple aspect ratios (21:9, 16:9, 4:3, 1:1, 3:4, 9:16, 9:21)
  • Configurable safety tolerance (0-6 scale)
  • Support for both JPEG and PNG output formats
  • Seed support for reproducible results

Advanced Finetuning System:

  • Complete finetuning workflow integration
  • Model customization with multiple training modes
  • Training mode selection (character/product/style/general)
  • Inference with adjustable strength control
  • Support for both full and LoRA finetuning
  • Comprehensive parameter control

Reliability & Usability:

  • Intelligent retry logic with exponential backoff
  • Graceful error handling with informative messages
  • Automatic config file discovery
  • Session-based HTTP requests for better performance
  • Detailed progress reporting and status updates

📋 Requirements

  • ComfyUI (latest version recommended)
  • Black Forest Labs API key
  • Python packages (auto-installed with ComfyUI):
    • requests
    • Pillow (PIL)
    • numpy
    • torch

🚀 Installation

  1. Create the node directory:

    ComfyUI/custom_nodes/flux_pro_integrative/

  2. Download the files:

    • __init__.py
    • flux_pro_integrative.py
    • config.ini (template)

  3. Set up your API key:

    Create or edit config.ini in the node directory:

    [API]
X_KEY=your_actual_api_key_here

  4. Restart ComfyUI

The node will automatically appear as "🎨 Flux Pro Integrative" in the BFL/Flux Pro category.

🔑 Getting Your API Key

  1. Create Account: Visit Black Forest Labs API Portal
  2. Generate Key: Access your dashboard and create a new API key
  3. Add to Config: Copy the key to your config.ini file

Useful Links:

💰 Pricing (Current Rates)

Image Generation

  • FLUX 1.1 Pro Ultra: $0.06 per image - Best for photo-realistic images at 2K+ resolution
  • FLUX 1.1 Pro Ultra Finetuned: $0.07 per image - Using your custom models
  • FLUX 1.1 Pro: $0.04 per image - Efficient for large-scale generation
  • FLUX.1 Pro: $0.05 per image - Original pro model
  • FLUX.1 Dev: $0.025 per image - Distilled model for development

Finetuning Training

  • Short ($2): < 150 steps - Quick exploration and testing
  • Medium ($4): 150-500 steps - Standard use cases
  • Long ($6): > 500 steps - Complex tasks requiring precision

🎛️ Node Interface

Core Parameters

Mode Selection:

  • generate: Create new images from prompts
  • finetune: Train custom models on your data
  • inference: Generate images using trained models

Generation Settings:

  • prompt: Text description of desired image (supports multiline)
  • ultra_mode: Enable Ultra mode for higher quality (default: True)
  • aspect_ratio: Choose from 7 ratios (default: 16:9)
  • safety_tolerance: Content filter strength 0-6 (default: 6)
  • output_format: PNG or JPEG output
  • raw: Skip safety filters for natural results (Ultra only)
  • seed: Set specific seed for reproducibility (-1 for random)

Finetuning Parameters

Training Setup:

  • finetune_zip: Path to training data ZIP file
  • finetune_comment: Description of your model (required)
  • trigger_word: Word to activate your concept (default: "TOK")
  • finetune_mode: Training type (character/product/style/general)

Advanced Training:

  • iterations: Training steps (100-2000, default: 300)
  • learning_rate: Training rate (default: 0.00001)
  • captioning: Auto-generate image descriptions (default: True)
  • priority: Training priority (speed/quality)
  • finetune_type: Full model or LoRA training
  • lora_rank: LoRA complexity (8-128, default: 32)

Inference Settings:

  • finetune_id: ID of your trained model
  • finetune_strength: Effect intensity (0.1-2.0, default: 1.2)

📖 Usage Guide

Basic Image Generation

  1. Add "🎨 Flux Pro Integrative" node to workflow
  2. Set mode to "generate"
  3. Enter your prompt
  4. Configure settings (aspect ratio, safety, etc.)
  5. Connect to Preview Image node

Example prompt:

A majestic mountain landscape at sunset, golden hour lighting, 
photorealistic, highly detailed, dramatic clouds

Training Custom Models

Step 1: Prepare Training Data

  • Collect 5-20 high-quality images (JPG, PNG, WebP)
  • Optionally add .txt files with descriptions
  • Create ZIP file with all training data

Step 2: Start Training

  • Set mode to "finetune"
  • Provide ZIP file path
  • Set descriptive comment
  • Choose appropriate finetune_mode:
    • character: For people, fictional characters
    • product: For objects, items, brands
    • style: For artistic styles, aesthetics
    • general: For concepts, scenes, general use

Step 3: Monitor Training

  • Node returns finetune ID
  • Training happens on BFL servers
  • Check status via API portal

Using Trained Models

  1. Set mode to "inference"
  2. Enter your finetune_id
  3. Include trigger_word in prompt
  4. Adjust finetune_strength as needed

Example inference prompt:

TOK character wearing a red dress in a garden, 
professional photography, portrait

🎯 Best Practices

Finetuning Success Tips

Image Quality:

  • Use high-resolution, clear images (1024px+ recommended)
  • Ensure consistent lighting and quality
  • Avoid blurry or heavily compressed images

Training Data:

  • For characters: Single person per image, various poses/angles
  • For products: Different angles, lighting conditions
  • For styles: Consistent artistic elements across images

Parameter Tuning:

  • Start with default settings
  • Increase iterations for complex concepts (500-1000)
  • Adjust finetune_strength during inference:
    • 0.8-1.0: Subtle influence
    • 1.2-1.5: Strong effect (default range)
    • 1.6-2.0: Very strong (may cause artifacts)

Generation Optimization

Prompt Engineering:

  • Be specific and descriptive
  • Include style keywords (photorealistic, artistic, etc.)
  • Mention lighting, composition, quality descriptors

Technical Settings:

  • Use Raw mode for photorealistic content
  • Higher safety_tolerance for creative freedom
  • PNG for detailed images, JPEG for photography

🔧 Troubleshooting

Common Issues

"Node not properly initialized"

  • Check if config.ini exists in node directory
  • Verify API key format: X_KEY=your_key under [API] section
  • Ensure no extra spaces around the API key

"Config file not found" The node searches these locations:

  1. ComfyUI/custom_nodes/flux_pro_integrative/config.ini
  2. ComfyUI/config.ini
  3. Current working directory

API Errors

  • HTTP 401: Invalid or expired API key
  • HTTP 429: Rate limiting (wait and retry)
  • HTTP 500: Server error (retry later)

Network Issues

  • Check internet connection
  • Verify firewall/proxy settings
  • Try again after a few minutes

Error Recovery

The node includes comprehensive error handling:

  • Returns blank image with error message on failure
  • Automatic retry logic for transient errors
  • Detailed console logging for debugging
  • Graceful degradation when services unavailable

🏗️ Technical Architecture

Class Structure

  • ConfigManager: Handles configuration loading and validation
  • FluxAPIClient: Manages all API interactions with robust error handling
  • FluxProIntegrative: Main ComfyUI node class with user interface
  • FluxAPIError: Custom exception for API-related errors

Key Improvements

  • Session Management: Reuses HTTP connections for better performance
  • Smart Retries: Exponential backoff for failed requests
  • Type Safety: Full type hints for better code reliability
  • Modular Design: Easy to extend and maintain

🤝 Contributing

Contributions are welcome! Please feel free to submit:

  • Bug reports with detailed reproduction steps
  • Feature requests with use case descriptions
  • Code improvements via pull requests
  • Documentation enhancements

📝 Changelog

v2.0.0 (Current)

  • Complete architectural rewrite
  • Enhanced error handling and user experience
  • Modern ComfyUI compatibility
  • Improved configuration management
  • Better logging and debugging tools

v1.0.0 (Legacy)

  • Initial release with basic API integration

🙏 Acknowledgements

  • Black Forest Labs for their powerful FLUX API
  • ComfyUI Community for the excellent node framework
  • Contributors who help improve this project

Note: This node requires an active Black Forest Labs API subscription. Please review their pricing and terms of service before use.