ComfyUI Flux Node - Flux Pro 1.1 Ultra & Raw with Finetuning

A custom node for ComfyUI that integrates with Black Forest Labs FLUX API, providing access to FLUX's image generation and finetuning capabilities.

Features

Core Features:

• Support for FLUX 1.1 [pro] regular and Ultra modes
• Optional Raw mode for more natural-looking images
• Multiple aspect ratios support
• Configurable safety tolerance
• Support for both JPEG and PNG output formats
• Seed support for reproducible results

Additional Capabilities:

• Complete finetuning system integration
• Model customization options
• Training mode selection (character/product/style/general)
• Inference with adjustable strength
• Comprehensive error handling

Requirements

• ComfyUI
• Black Forest Labs API key
• Python packages:
  • requests
  • Pillow
  • numpy
  • torch

Installation

1. Create a folder in your ComfyUI/custom_nodes/ directory named ComfyUI_Flux_1.1_RAW_API
2. Copy the files from this repository into that folder
3. Create a config.ini file in the same directory with your FLUX API key

Example config.ini:

[API]
X_KEY=your_api_key_here
BASE_URL=https://api.bfl.ai/

Useful Links

• Main Website: Black Forest Labs
• API Portal: https://api.us1.bfl.ai/auth/login

Getting Started

1. Create an account on the API Portal
2. Generate your API key from the dashboard
3. Create a config.ini file in your node directory with your API key (see Installation section)

Pricing

Finetuning Training

• Short ($2): < 150 steps - For fast exploration
• Medium ($4): 150-500 steps - For standard use cases
• Long ($6): > 500 steps - For difficult tasks and extreme precision

Image Generation

• FLUX 1.1 Pro Ultra Finetuned: $0.07 per image
• FLUX 1.1 Pro Ultra: $0.06 per image - Best for photo-realistic images at 2k resolution
• 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

Usage

Node Parameters

• prompt: Text prompt describing the desired image
• ultra_mode: Enable Ultra mode for higher resolution output
• aspect_ratio: Choose from multiple aspect ratios:
  • 21:9 (Ultrawide)
  • 16:9 (Widescreen)
  • 4:3 (Standard)
  • 1:1 (Square)
  • 3:4 (Portrait)
  • 9:16 (Vertical)
  • 9:21 (Tall)
• safety_tolerance: 0-6 (0 being most strict, 6 being least strict)
• output_format: Choose between JPEG and PNG
• raw: Enable Raw mode for less processed, more natural-looking images
• seed: Optional seed for reproducible results (-1 for random)

Mode Differences

Regular Mode

• Resolution based on aspect ratio (up to 1440px)
• Standard image processing

Ultra Mode

• Higher resolution output (up to 4MP)
• Support for Raw mode
• Advanced image processing

Basic Workflow

1. Add the "Flux Pro 1.1 Ultra & Raw with Finetuning" node to your workflow
2. Set mode to "generate"
3. Enter your prompt
4. Configure desired settings (aspect ratio, safety tolerance, etc.)
5. Connect to a Preview Image node to see results

Finetuning

1. Prepare your training data:

   • Create a folder with 5-20 images (JPG, JPEG, PNG, or WebP)
   • Optionally add text descriptions (same filename but .txt extension)
   • Compress the folder into a ZIP file

2. Training:

   • Set mode to "finetune"
   • Provide path to your ZIP file
   • Set finetune parameters:
     • finetune_comment: Description of your model
     • finetune_mode: character/product/style/general
     • trigger_word: Word to reference your concept (default: "TOK")
     • Adjust other parameters as needed

3. Using Finetuned Models:

   • Set mode to "inference"
   • Input your finetune_id
   • Use the trigger word in your prompt
   • Adjust finetune_strength if needed (default: 1.2)

Parameters

Generation Parameters

• mode: generate/finetune/inference
• prompt: Text description of desired image
• ultra_mode: Enable/disable ultra mode
• aspect_ratio: 21:9, 16:9, 4:3, 1:1, 3:4, 9:16, 9:21
• safety_tolerance: 0-6
• output_format: jpeg/png
• raw: Enable raw mode for extra realism
• seed: Set specific seed (-1 for random)

Finetuning Parameters

• finetune_zip: Path to training data ZIP
• finetune_comment: Model description
• finetune_id: ID for inference mode
• trigger_word: Concept reference word
• finetune_mode: character/product/style/general
• iterations: Training steps (min: 100)
• learning_rate: Training rate (default: 0.00001)
• captioning: Auto-caption images
• priority: speed/quality
• finetune_type: full/lora
• lora_rank: 16 or 32
• finetune_strength: 0-2 (default: 1.2)

Best Practices

Finetuning Tips

• Use high-quality training images
• For character training, avoid multiple characters in single images
• For style training, try higher finetune_strength values
• If concept isn't showing strongly, increase finetune_strength
• If seeing artifacts, decrease finetune_strength

Image Generation Tips

• Be specific in your prompts
• Use raw mode for more photorealistic results
• Adjust safety_tolerance based on your needs
• Consider aspect ratio for your use case

Troubleshooting

Common issues:

1. "Error: X_KEY not found in section API of config file"

   • Check if your config.ini file exists and is properly formatted
   • Verify the API key is correctly copied without extra spaces
   • Make sure the file is in the correct directory

2. "404 Not Found" error

   • Verify your API key is valid and active
   • Check if you have access to the FLUX 1.1 API
   • Make sure your account is in good standing

Error Handling

The node includes comprehensive error handling:

• Network errors return a blank image
• Invalid API keys are caught and reported
• Missing files generate appropriate errors
• Training/inference errors are logged with details

Contributing

Feel free to submit issues and enhancement requests!

Acknowledgements

• FLUX API team for their powerful image generation capabilities
• ComfyUI community for the node framework