PhotoRX - Film Grain Custom Nodes for ComfyUI

A physics-based film grain synthesis plugin that brings authentic film grain to your ComfyUI workflows. Based on the inhomogeneous Boolean model, it accurately emulates the statistical properties of real photographic film.

Features

• Physics-Based Rendering: Uses the proven Boolean model from computational photography research
• Two Renderers: Fast analytical renderer (~100-500x faster) or Monte Carlo for reference quality
• Film Profiles: Pre-configured profiles for popular film stocks
• Batch Processing: Works with ComfyUI's batch system
• Reproducible Results: Seed-based randomization for consistent outputs

Installation

Manual Installation

1. Clone or download this repository to your ComfyUI custom nodes directory:

   cd ComfyUI/custom_nodes/
   git clone https://github.com/yourusername/ComfyUi-PhotoRX.git
   

2. Install dependencies:

   pip install -r ComfyUi-PhotoRX/requirements.txt
   

3. Restart ComfyUI

Using ComfyUI Manager

If available, search for "PhotoRX" in the Custom Nodes Manager and install.

Usage

After installation, restart ComfyUI. The nodes will appear under the PhotoRX category:

Film Grain (Basic)

Simple interface for quick grain application.

Inputs:

• image - Input image tensor
• film_profile - Select a film stock (default: Tri-X 400)
• black_white - Convert to grayscale before applying grain
• seed - Random seed for reproducibility
• zoom - Output resolution multiplier (optional, default: 1.0)

Outputs:

• image - Image with film grain applied

Film Grain (Advanced)

Full control over all grain parameters.

Inputs:

• image - Input image tensor
• film_profile - Base film stock for defaults
• renderer - "fast" or "monte_carlo"
• black_white - Convert to grayscale before applying grain
• seed - Random seed for reproducibility

Optional Parameters:

• mu_r - Mean grain radius (default: 0.07)
• sigma_r - Grain radius standard deviation (default: 0.025)
• filter_sigma - Gaussian filter sigma in output pixels (default: 0.8)
• mc_samples - Monte Carlo samples per pixel (default: 100)
• zoom - Output resolution multiplier (default: 1.0)

Outputs:

• image - Image with film grain applied

Profile Info (Helper)

Displays information about a selected film profile. Useful for exploring available options.

Film Profiles

Black & White Films

| Profile | Description | |---------|-------------| | tri-x | Kodak Tri-X 400 - Classic high-contrast B&W | | hp5-plus | Ilford HP5 Plus - Versatile all-purpose film | | t-max-100 | Kodak T-Max 100 - Fine grain, sharp | | t-max-400 | Kodak T-Max 400 - Fine grain at higher speed | | t-max-p3200 | Kodak T-Max P3200 - Ultra high speed | | delta-100 | Kodak Portra 400NC - Smooth, fine grain | | delta-400 | Kodak Portra 400NC - Classic portrait film | | pan-f-plus | Fujifilm Acros 100 - Ultra-fine grain | | acros | Fujifilm Acros 100 - Modern fine grain B&W |

Color Films

| Profile | Description | |---------|-------------| | portra-160 | Kodak Portra 160 - Fine grain portrait film | | portra-400 | Kodak Portra 400 - Classic portrait film | | ektar-100 | Kodak Ektar 100 - Vibrant colors, fine grain | | superia-400 | Fujifilm Superia 400 - Everyday color negative | | gold-200 | Kodak Gold 200 - Warm consumer film | | ultramax-400 | Kodak Ultramax 400 - High-speed consumer |

Example Workflow

Load Image → Film Grain (Basic) → Save Image

Or for more control:

Load Image → Film Grain (Advanced) → Adjust Parameters → Save Image

Performance Notes

• Fast Renderer: Typically completes in <1 second for HD images
• Monte Carlo Renderer: Can take 30-60 seconds depending on sample count and image size
• First Monte Carlo run includes JIT compilation overhead

Troubleshooting

"Module not found" errors

Ensure all dependencies are installed:

pip install numpy numba Pillow scipy

Slow first render with Monte Carlo

This is expected - Numba JIT compiles on first use. Subsequent renders will be faster.

Grain looks too strong/weak

Adjust the mu_r parameter (grain size) or use the zoom parameter to scale the effect.

License

MIT License - See LICENSE file for details.

Credits

Based on the GrainRX film grain synthesis library. The Boolean model implementation follows research from computational photography literature.

---

For issues and feature requests, please visit the GitHub repository.