ComfyUI_IC-Light-v2_fal

Features

• Re-lighting photos and changing their backgrounds based on a given description
• Use your own native ComfyUI mask as input (optional)
• Adjustable generation parameters (seed, lighting direction, denoise strength, etc.)
• Multiple image generation in a single request

Prerequisites

• ComfyUI installed and set up
• fal.ai account and API key

Installation

You can install this custom node using the ComfyUI Manager or manually.

Install via ComfyUI Manager

1. Install ComfyUI Manager if you haven't already.
2. Open ComfyUI and click the "Manager" button.
3. Search for "ComfyUI_IC-Light-v2_fal" in the custom nodes section.
4. Click "Install" to automatically download and install the custom nodes.

Install manually

1. Clone this repository into your ComfyUI's custom_nodes directory:

cd /path/to/ComfyUI/custom_nodes
git clone https://github.com/a-und-b/ComfyUI_IC-Light-v2_fal.git

2. Navigate to the cloned directory:

cd ComfyUI_IC-Light-v2_fal

3. Install the required dependencies:

pip install -r requirements.txt

4. Configure your API key:

   • Copy config.ini.template to config.ini in the root directory of this custom node (ComfyUI/custom_nodes/ComfyUI_IC-Light-v2_fal).
   • Add your fal.ai API key to the config.ini file

5. Restart ComfyUI if it's already running

Usage

After installation, you'll find the "IC-Light V2 (fal.ai)" node in the ComfyUI interface. Connect an input image and an optional mask to change the subject's background and lighting based on your prompt.

Parameters

• image: Input image to be relit
• prompt: Text prompt describing the desired lighting or background
• negative_prompt: Negative prompt to guide the model away from certain concepts
• num_inference_steps: Number of denoising steps (impacts generation time and quality)
• guidance_scale: Higher values enforce the prompt more strictly
• seed: Random seed for reproducible results
• initial_latent: Lighting direction - None, Left, Right, Top, or Bottom
• enable_hr_fix: Enable high-resolution fix for better quality
• lowres_denoise: Strength for low-resolution pass
• highres_denoise: Strength for high-resolution pass (used with HR fix)
• hr_downscale: Downscale factor for high-resolution pass
• num_images: Number of variations to generate
• cfg: Classifier-Free Guidance scale for generation
• output_format: Output image format (JPEG or PNG)
• mask (optional): Mask to target specific areas of the image

Example Workflow

You can load the JSON file from the example_workflows folder directly into ComfyUI (e.g., by dragging and dropping the file onto the ComfyUI window or using the 'Load' button).

This workflow incorporates custom nodes from the following widely used extension packages:

• ComfyUI-Easy-Use
• ComfyUI-KJNodes

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgements

• lllyasviel for the IC-Light model
• comfyanonymus for bringing us ComfyUI
• behzad kazemi for the Photo used in the example workflow
• fal.ai for providing the API endpoint

Verifiziert 11. Sept.