Eric_Image_Processing_Nodes

AdaptiveImageEnhancement

Intelligent image enhancement that analyzes image content and applies appropriate enhancement techniques automatically Features: - Automatic content analysis (documents, photos, artwork) - Adaptive enhancement pipeline - Quality assessment and adjustment - Multi-stage processing with fallbacks

Eric'S Image Processing/Enhancement · 1 input · 8 parameters · 1 output

AdvancedFilmGrainNode

Complete film grain processing with automatic analysis: Input Parameters: - processingmode: Auto-analyze or manual specification - graintype: Manual grain type selection - denoisingmethod: GPU wavelet, stationary, CPU, or adaptive - preservationlevel: How much grain to preserve (0.0-1.0) - wavelettype: Wavelet family selection - thresholdingmethod: BayesShrink, SureShrink, VisuShrink, manual - usegpu: Enable GPU acceleration - multistageprocessing: Enable multi-stage processing pipeline - edgepreservation: Preserve edge information - grainenhancement: Enhance grain structure Outputs: - processedimage: Denoised image - grainanalysis: Detailed grain analysis report - processinginfo: Processing method and performance information

Eric'S Nodes/Advanced Processing · 1 input · 15 parameters · 3 outputs

AutoDenoiseComparisonNode

Auto-Denoise Comparison Node: Compare multiple state-of-the-art methods Processes the same image with different Auto-Denoise methods and provides visual comparison with quality metrics. Methods compared: - Noise2Void (self-supervised) - Deep Image Prior (unsupervised) - Original image for reference

Eric'S Image Processing/Auto Denoise · 1 input · 5 parameters · 4 outputs

AutoDenoiseNode

Auto-Denoise Node: Intelligent denoising method selection and processing Automatically analyzes the image and selects the optimal denoising method: - Noise2Void for high-noise images - Deep Image Prior for low-complexity images Features: - No clean reference images required - Self-supervised learning - Automatic method selection - GPU acceleration support

Eric'S Image Processing/Auto Denoise · 1 input · 6 parameters · 2 outputs

BatchImageProcessing

Efficient batch processing node for applying the same enhancement to multiple images with optimized memory management Features: - Process multiple images with consistent parameters - Memory-efficient processing with garbage collection - Progress tracking and timing - Automatic parameter optimization per image - Fallback handling for individual image failures

Eric'S Image Processing/Batch · 1 input · 9 parameters · 1 output

BM3D Film Grain Denoising

BM3D-GT&AD Node for film grain denoising with resolution-aware parameter scaling This node is optimized for high-resolution images (1-20MP) with automatic parameter adjustment based on image resolution: Resolution Scaling: - <= 2MP: Full strength (scale factor 1.0) - 2-4MP: Reduced strength (scale factor 0.9) - 4-8MP: Further reduced (scale factor 0.75) - > 8MP: Minimal processing (scale factor 0.6) Grain Strength Presets (base sigma values): - ultra_light: 5.0 (ideal for high-res images) - light: 8.0 (subtle grain removal) - medium: 15.0 (balanced denoising) - heavy: 25.0 (strong grain removal) - custom: User-defined sigma (1.0-80.0) Patch Size Adaptation: - Automatically scales patch size based on image dimensions - Larger images get larger patches for better performance - Conservative scaling prevents overly aggressive processing

Eric'S Nodes/Advanced Enhancement · 1 input · 3 parameters · 2 outputs

BM3DComparisonNode

BM3D Profile Comparison Node: Compare different BM3D profiles Processes the same image with multiple BM3D profiles and provides visual comparison with quality metrics. Profiles compared: - np: Normal Profile (balanced) - refilter: Enhanced quality - high: Maximum quality - vn: Very noisy images

Eric'S Image Processing/BM3D · 1 input · 4 parameters · 5 outputs

BM3DDeblurNode

BM3D Deblurring Node: Combined deblurring and denoising Performs joint deblurring and denoising using BM3D algorithm. Requires a point spread function (PSF) to model the blur. Perfect for: - Motion blur removal - Gaussian blur removal - Simultaneous blur and noise removal - Professional image restoration

Eric'S Image Processing/BM3D · 1 input · 6 parameters · 2 outputs

BM3DDenoiseNode

BM3D Denoising Node: State-of-the-art block-matching and 3D filtering Automatically handles both grayscale and color images using the BM3D algorithm, which excels at removing additive spatially correlated stationary Gaussian noise. Features: - Automatic grayscale/color detection - Multiple quality profiles - Automatic noise estimation - Professional-grade results

Eric'S Image Processing/BM3D · 1 input · 5 parameters · 2 outputs

ComprehensiveComparison

Comprehensive Image Processing Comparison Compare multiple processing methods side-by-side: AI METHODS: - SCUNet: Realistic restoration with blind denoising - SwinIR: Transformer-based precision processing - Real-ESRGAN: Real-world photo enhancement TRADITIONAL METHODS: - Wavelet Denoising: Mathematical noise reduction - Frequency Enhancement: Spectral domain processing - Bilateral Filtering: Edge-preserving smoothing COMPARISON FEATURES: - Side-by-side visual results - Quantitative quality metrics - Processing time comparison - Memory usage analysis - Method recommendations ANALYSIS METRICS: - PSNR (Peak Signal-to-Noise Ratio) - SSIM (Structural Similarity) - Processing time per method - Memory efficiency - Visual quality assessment Perfect for: - Method evaluation and selection - Research and benchmarking - Understanding method strengths - Quality comparison studies

1 input · 6 parameters · 3 outputs

DeepImagePriorNode

Deep Image Prior Node: Unsupervised restoration using CNN architecture Uses the implicit bias of CNN architectures for image restoration. No training data required - the network structure acts as a prior. Excellent for: - Image restoration without training data - Denoising with unknown noise models - Super-resolution and inpainting - Preserving image structure

Eric'S Image Processing/Auto Denoise · 1 input · 5 parameters · 2 outputs

DiffBIRRestorationNode

Diffusion-based blind restoration using DiffBIR.

Eric'S Image Processing/Transformer Enhancements · 1 output

DnCNNDenoiseNode

Classic deep learning denoiser from 2017 that still performs excellently. 1. dncnn25 (Default, Recommended) - Noise level: σ=25 - Best for moderate noise - Grayscale processing 2. dncnn15 - Noise level: σ=15 - Best for light noise - Grayscale processing 3. dncnn50 - Noise level: σ=50 - Best for heavy noise - Grayscale processing 4. dncnn3 - Blind denoising (any noise level) - Grayscale processing - More flexible but slightly less accurate 5. dncnncolorblind - Blind denoising for RGB images - Color processing - Most versatile - ✅ Fast processing needed - ✅ Film grain or moderate noise - ✅ You know the approximate noise level - ✅ Smaller file sizes preferred

Eric'S Image Processing/Pre Trained Denoisers · 1 input · 1 parameter · 2 outputs

FGANNFilmGrainDenoiseNode

FGA-NN Film Grain Denoising Node Modern neural network approach for film grain removal using: - Multi-scale feature extraction - Auto-regressive grain parameter estimation - CBAM attention mechanisms - Residual learning for noise prediction Perfect for: - Scanned film footage restoration - Digital film grain removal - Grain characteristic analysis - Professional post-production workflows

Eric'S Image Processing/Film Grain · 1 input · 2 parameters · 2 outputs

FilmGrainAnalysis

Film grain analysis node that provides detailed grain characteristics without processing the image

Eric'S Image Processing/Film Grain · 1 input · 2 parameters · 2 outputs

FilmGrainAnalysisNode

Diagnostic tool for grain analysis: Outputs: - graintype: Detected grain type - analysissummary: Detailed analysis metrics - recommendations: Suggested processing parameters

Eric'S Image Processing/Film Grain · 1 input · 3 parameters · 2 outputs

FilmGrainProcessing

1. Automatic Processing: - Load image with grain/noise - Add Advanced Film Grain Processing (Eric) node - Set processingmode to "autoanalyze" - Adjust preservationlevel (0.0-1.0) - Enable usegpu for faster processing - Process to get automatically optimized results 2. Manual Processing for Specific Grain Types: - Set processingmode to "manualspecify" - Choose appropriate graintype: - Authentic Film: For real film grain (use preservation 0.7-0.9) - Digital Noise: For high-ISO camera noise (use preservation 0.2-0.5) - Simulated Overlay: For artificial grain effects (use preservation 0.0-0.3) - Select denoisingmethod (gpuwavelet recommended) - Enable multistageprocessing for best results 3. GPU-Accelerated Wavelet Only: - Add GPU Wavelet Denoising (Eric) node - Set usegpu to True - Choose wavelettype and thresholdingmethod - Enable usestationary for translation-invariant processing 4. Grain Analysis Only: - Add Film Grain Analysis (Eric) node - Enable detailedanalysis for comprehensive metrics - Review grain type detection and recommendations

Eric'S Image Processing/Film Grain · 1 input · 6 parameters · 2 outputs

FrequencyEnhancementPresets

Preset configurations for common frequency enhancement scenarios

Eric'S Image Processing/Frequency Enhancement · 1 output

GPUWaveletDenoiseNode

Standalone GPU-accelerated wavelet denoising: Features: - Direct GPU wavelet processing - Stationary wavelet transform option - Automatic CPU fallback - Memory management - Performance reporting

Eric'S Nodes/Denoising · 1 input · 8 parameters · 2 outputs

HomomorphicFilter

Homomorphic filtering for illumination and reflectance separation Excellent for: - Correcting uneven lighting in scanned photos - Shadow removal and highlight recovery - Normalizing illumination across image regions - Scientific imaging with lighting variations Performance: Fast (FFT-based), dramatic lighting improvements

Eric'S Image Processing/Frequency Enhancement · 1 input · 5 parameters · 1 output

ImageQualityAssessment

Comprehensive image quality assessment node for evaluating enhancement results Provides: - No-reference quality metrics (doesn't need original) - Full-reference metrics (compares to original) - Content analysis metrics - Enhancement effectiveness scores

Eric'S Image Processing/Analysis · 2 inputs · 4 parameters · 2 outputs

LB-CLAHE

Input: Low contrast image Expected: Enhanced contrast with preserved colors

Eric'S Nodes/Advanced Enhancement · 1 input · 3 parameters · 2 outputs

LightweightCNNDenoiseNode

Lightweight Progressive CNN Denoising Node Efficient neural network approach for film grain removal using: - Dense blocks for comprehensive feature extraction - Progressive residual fusion (shallow → deep features) - Lightweight attention mechanisms - Minimal parameters (~500KB) for fast inference Perfect for: - Real-time denoising applications - Resource-constrained environments - Batch processing workflows - Quick iterative editing

Eric'S Image Processing/Film Grain · 1 input · 2 parameters · 2 outputs

MemoryOptimization

Memory Management and Optimization Node Provides intelligent memory management for image processing workflows: - Automatic tile size optimization - Memory usage monitoring - GPU memory cleanup - Processing optimization recommendations Essential for: - Large image processing (>4MP) - GPU memory management - Workflow optimization - Preventing out-of-memory errors

1 input · 4 parameters · 2 outputs

Multi-Scale Retinex

- 512x512: Smaller scales for faster processing - 2048x2048: Proportionally larger scales maintain quality - 3000x2000: Scales capped at reasonable maximums

Eric'S Nodes/Advanced Enhancement · 1 input · 7 parameters · 2 outputs

MultiscaleFFTEnhancement

Multi-scale frequency domain enhancement using Laplacian pyramid Excellent for: - Sophisticated detail enhancement across multiple scales - Professional photo retouching - Scientific image analysis requiring scale-specific enhancement - Replacing traditional unsharp masking with superior results Performance: Medium speed, excellent quality

Eric'S Image Processing/Frequency Enhancement · 1 input · 7 parameters · 1 output

NAFNetDenoiseNode

State-of-the-art 2022 architecture with no activation functions - more efficient! 1. nafnet-width32 (Default, Recommended) - Lighter, faster model - 2M parameters - Best for speed 2. nafnet-width64 - Balanced quality/speed - 8M parameters - Best all-around 3. nafnet-sidd - Trained on real-world noise (SIDD dataset) - Best for camera noise, ISO noise - Handles complex real-world scenarios - ✅ Best quality needed - ✅ Real-world camera noise - ✅ Complex noise patterns - ✅ Latest technology preferred

Eric'S Image Processing/Pre Trained Denoisers · 1 input · 2 parameters · 2 outputs

Noise2VoidNode

Noise2Void Node: Self-supervised single image denoising Revolutionary approach that doesn't require clean reference images. Works by training on blind spots in the same image. Perfect for: - Single noisy images - Unknown noise characteristics - Real-world noise patterns - Film grain and digital noise

Eric'S Image Processing/Auto Denoise · 1 input · 5 parameters · 2 outputs

NoiseDABatchNode

Batch processing node for Noise-DA models

Eric'S Nodes/AI Models · 1 input · 4 parameters · 2 outputs

NoiseDANode

ComfyUI Node for Noise-DA pretrained models

Eric'S Nodes/AI Models · 1 input · 6 parameters · 2 outputs

NonLocalMeans

- h Parameter: Filtering strength (auto-estimated if 0) - 0.1-1.0: Ultra-light denoising for high-quality images - 1.0-5.0: Subtle denoising, preserve fine details - 5.0-15.0: Balanced denoising (recommended) - 15.0-30.0: Strong denoising - 30.0+: Very strong, may blur details - Patch Size: 5-7 for fine textures, 7-11 for coarse textures - Search Window: 11-15 standard, 17-50 for maximum quality

Eric'S Image Processing/Denoising · 1 input · 9 parameters · 1 output

NonLocalMeansSimple

Simplified Non-Local Means node with automatic parameter selection

Eric'S Image Processing/Denoising · 1 input · 2 parameters · 1 output

PhasePreservingEnhancement

Phase-preserving frequency domain enhancement Excellent for: - Natural-looking detail enhancement - Sharpening without ringing artifacts - Preserving image character while boosting details - Scientific and medical image enhancement Performance: Fast (FFT-based), natural results

Eric'S Image Processing/Frequency Enhancement · 1 input · 5 parameters · 1 output

ProfessionalRestorationPipeline

Professional Multi-Stage Restoration Pipeline Combines multiple advanced processing methods in an intelligent sequence: PIPELINE STAGES: 1. Pre-processing: Noise reduction and artifact removal 2. Primary Restoration: AI-based enhancement (SCUNet/SwinIR/Real-ESRGAN) 3. Frequency Enhancement: Detail recovery and sharpening 4. Color Optimization: Perceptual color enhancement 5. Post-processing: Final refinement and quality assurance INTELLIGENT ADAPTATION: - Automatically adjusts pipeline based on image characteristics - Skips unnecessary stages for optimal efficiency - Applies appropriate strength levels for each stage - Monitors quality at each step to prevent over-processing PROFESSIONAL FEATURES: - Quality-guided processing strength - Multi-scale analysis and processing - Edge-preserving enhancement - Noise-adaptive filtering - Color space optimization Perfect for: - Professional photo restoration - High-quality archival processing - Complex degradation scenarios - Demanding quality requirements

1 input · 8 parameters · 2 outputs

RealBM3DDeblurNode

Real BM3D Deblurring Node: GPU-accelerated deblurring with real BM3D Uses the actual BM3D deblurring algorithm with CUDA acceleration. Performs joint deblurring and denoising in a single step.

Eric'S Image Processing/Real BM3D · 1 input · 8 parameters · 2 outputs

RealBM3DNode

Real BM3D Node: GPU-accelerated true BM3D implementation Uses the actual BM3D library with CUDA acceleration for maximum performance and quality. This is the real BM3D algorithm, not an approximation. Features: - True BM3D algorithm implementation - CUDA GPU acceleration - Multiple BM3D profiles (np, refilter, high, vn, etc.) - Automatic noise estimation - Both grayscale and color processing - Professional denoising results

Eric'S Image Processing/Real BM3D · 1 input · 6 parameters · 2 outputs

RestormerRestorationNode

Restormer restoration with optional sharpness boosting.

Eric'S Image Processing/Transformer Enhancements · 1 output

RichardsonLucy

- No code changes needed - Just need correct parameters for AI images - Default settings are correct for real camera blur

Eric'S Image Processing/Restoration · 1 output

RichardsonLucyGPU

Compact description unavailable for this node.

Eric'S Image Processing/Restoration · 1 output

RichardsonLucySimple

Simplified Richardson-Lucy node with preset configurations

Eric'S Image Processing/Restoration · 1 input · 4 parameters · 1 output

SCUNetBatchRestoration

SCUNet Batch Processing for Multiple Images Efficiently process multiple images with shared model loading Ideal for: - Processing entire photo collections - Batch restoration workflows - Video frame processing Performance: Optimized for batch processing with model reuse

1 input · 2 parameters · 1 output

SCUNetRestoration

SCUNet (Swin-Conv-UNet) Deep Learning Image Restoration State-of-the-art restoration for: - Film grain and digital noise removal - JPEG compression artifact reduction - Motion blur and defocus blur correction - Mixed degradation restoration - Old photo restoration - Low-light image enhancement Performance: GPU-accelerated, excellent quality on complex degradations Memory: Optimized with tiling for large images

1 output

Smart Sharpening

Input: Soft image Expected: Intelligently sharpened image If you still see this error pattern: error: (-215:Assertion failed) dims 0 in function 'cv::Mat::locateROI' Action: 1. Identify the specific OpenCV function call 2. Add array = np.ascontiguousarray(array) before the call 3. If after color conversion or slicing, add contiguous check np.ascontiguousarray() has minimal overhead: - If array is already contiguous: 0μs (just returns same array) - If copy needed: 1ms for 1080p (memory copy operation) Trade-off: Slight overhead for guaranteed compatibility. Two OpenCV packages installed simultaneously: opencv-python 4.12.0.88 opencv-python-headless 4.12.0.88 Option A: Keep opencv-python (GUI support) bash pip uninstall opencv-python-headless

Eric'S Nodes/Advanced Enhancement · 1 input · 4 parameters · 2 outputs

SmartWorkflowNode

Smart Workflow - Intelligent Image Processing Selection Automatically analyzes your image and selects the optimal processing method: AI PROCESSORS: - SCUNet: Best for realistic image restoration and blind denoising - SwinIR: Excellent for classical super-resolution and clean images - Real-ESRGAN: Great for real-world images with complex degradation ANALYSIS FEATURES: - Noise level detection - Blur assessment - Compression artifact identification - Resolution analysis - Content type recognition SMART SELECTION: - High noise → SCUNet (blind denoising specialist) - Clean images → SwinIR (transformer precision) - Real-world photos → Real-ESRGAN (practical restoration) - Mixed issues → Multi-stage processing Perfect for: - Users unsure which method to use - Batch processing with varied image types - Professional workflows requiring consistent results

1 input · 6 parameters · 2 outputs

StationaryWaveletDenoise

Stationary (translation-invariant) wavelet denoising node Excellent for: - Images with important structures at different positions - Reducing shift-sensitivity artifacts - Higher quality denoising with less artifacts Performance: Slower than regular wavelets but better quality

Eric'S Image Processing/Denoising · 1 input · 8 parameters · 1 output

SwinIRBatch

SwinIR Batch Processing for Multiple Images Efficiently process multiple images with SwinIR transformer Ideal for: - Batch photo restoration - Consistent processing across image sets - Professional workflow integration

1 input · 3 parameters · 1 output

SwinIRRestoration

SwinIR Transformer-based Image Restoration State-of-the-art restoration using Swin Transformer architecture: - 67% parameter reduction compared to CNN methods - Up to 0.45dB PSNR improvement - Hierarchical feature extraction with local and global context - Residual Swin Transformer Blocks with shifted window attention Excellent for: - Classical image super-resolution - Image denoising (Gaussian, real noise) - JPEG compression artifact reduction - Professional photo restoration Performance: Transformer-based, excellent quality with efficient parameters

1 input · 4 parameters · 1 output

SwinIRSharpnessBoostNode

SwinIR restoration with user-adjustable sharpness boosting.

Eric'S Image Processing/Transformer Enhancements · 1 output

WaveletDenoise

Advanced wavelet denoising node with multiple threshold selection methods Excellent for: - Natural photographs with Gaussian noise - Scanned images with grain - General purpose denoising with edge preservation Performance: Very fast, runs on CPU, scales well with image size

Eric'S Image Processing/Denoising · 1 input · 8 parameters · 1 output

WienerFilter

- Blur Size: - 0.1-0.5: Ultra-fine sharpening for high-res images - 0.5-2.0: Light blur correction - 2.0-6.0: Moderate blur (recommended) - 6.0+: Heavy blur - K Value (Regularization): - 0.0001-0.001: Ultra-clean images, maximum sharpening - 0.001-0.01: Clean images, sharp results - 0.01-0.1: Moderate noise (recommended) - 0.1+: Noisy images, smooth results

Eric'S Image Processing/Restoration · 1 input · 7 parameters · 1 output

WienerFilterCompare

Compare different Wiener filter configurations side by side

Eric'S Image Processing/Restoration · 1 input · 1 parameter · 4 outputs

WienerFilterSNR

Simplified Wiener filter using Signal-to-Noise Ratio specification

Eric'S Image Processing/Restoration · 1 input · 2 parameters · 1 output