ComfyUI-IQA-Node

A comprehensive Image Quality Assessment (IQA) custom node collection for ComfyUI. This pack leverages PyIQA for deep learning-based metrics, OpenCV for classical computer vision metrics, and advanced Analysis Tools for detailed image inspection.

Features

• PyIQA Deep Image Analysis:

  • Access to a vast zoo of state-of-the-art IQA models (HyperIQA, MUSIQ, NIMA, LPIPS, FID, SSIM, etc.).
  • Smart Caching: Models are cached in memory to avoid reloading.
  • Batch Support: Processes batches with configurable aggregation.
  • Automatic Device Detection: Runs on CUDA if available.

• OpenCV & Analysis Tools:

  • Classical Metrics: Blur, Brightness, Contrast, Colorfulness, Noise.
  • Advanced Analysis: Color Harmony, Color Temperature, Defocus (FFT), Clipping, Entropy.
  • Visualizations: Histograms, Heatmaps, Color Wheels, Edge Maps.

• Logic & Visualization:

  • Filtering & Ranking: Filter or sort images based on quality scores.
  • Ensemble: Combine multiple scores.
  • Frontend Integration: Real-time score display on nodes.

Installation

1. Clone this repository into your ComfyUI/custom_nodes folder:

   cd ComfyUI/custom_nodes
   git clone https://github.com/yourusername/ComfyUI-IQA-Node.git

2. Install dependencies:

   pip install -r requirements.txt

   Note: This version uses a vendored version of pyiqa to prevent dependency conflicts (like downgrading transformers or numpy). All original pyiqa metrics are available except for qalign, which is disabled due to its strict requirement for an older version of transformers.

Usage

PyIQA Nodes

• IQA: PyIQA No-Reference: Evaluate aesthetics/quality (HyperIQA, MUSIQ, NIMA, etc.).
• IQA: PyIQA Full-Reference: Compare against reference (LPIPS, SSIM, PSNR, FID, etc.).

OpenCV Nodes

• IQA: Blur Estimation: Global blur detection (Laplacian/Tenengrad).
• IQA: Brightness & Contrast: Basic statistics.
• IQA: Colorfulness: Vividness metric.
• IQA: Noise Estimation: Wavelet-based noise estimation (MAD).
• IQA: Edge Density: Edge complexity.
• IQA: Saturation: Average/Max saturation.

Analysis Nodes

• Analysis: Blur Detection: Block-based blur analysis with heatmaps.
• Analysis: Color Harmony: Identifies color schemes (Complementary, Triadic) and displays color wheel.
• Analysis: Color Cast: Detects and visualizes unwanted color tints.
• Analysis: Color Temperature: Estimates Kelvin temperature and labels (Warm/Cool).
• Analysis: Clipping: Visualizes clipped highlights/shadows or saturation.
• Analysis: Defocus: FFT-based frequency analysis to detect defocus.
• Analysis: Edge Density: Detailed edge density analysis with maps.
• Analysis: Entropy: Measures information content/entropy (bits).
• Analysis: Noise Estimation: Variance-based noise mapping.
• Analysis: RGB Histogram: Renders RGB histograms for the batch.
• Analysis: Sharpness/Focus: Hybrid scoring (Laplacian + Tenengrad).

Logic Nodes

• IQA: Threshold Filter: Route images based on score.
• IQA: Batch Ranker: Sort images by score.
• IQA: Ensemble Scorer: Weighted average of scores.
• IQA: Score Normalizer: Scale/Invert scores.

Visualization Nodes

• IQA: Heatmap Visualizer: Apply colormaps to raw value maps.

Tips

• Use Analysis: Defocus or IQA: Blur Estimation to filter bad generations.
• Use Analysis: Color Harmony to verify prompt adherence (e.g., "teal and orange").
• Use IQA: Batch Ranker to pick the best image from a large batch.

Credits

• IQA-PyTorch for DL metrics.
• ComfyUI.
• ThatGlennD/ComfyUI-Image-Analysis-Tools for analysis logic.