nodes/EricRollei-AAA_Metadata_System
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
Files
main
EricRollei-AAA_Metadata_System/documentation/ReadMe Docs/MetadataStructureReference.md
T

12 KiB
Raw Permalink Blame History

Metadata Structure Reference

This document provides a comprehensive reference for the metadata structure used in the Metadata System for ComfyUI. It details the fields, their meanings, and their organization.

Overview

The metadata is organized into a structured hierarchy with four main sections:

  • basic: Essential information such as title, description, and keywords
  • analysis: Image analysis results including technical and aesthetic measures
  • ai_info: AI generation information including model, prompt, and workflow data
  • regions: Information about specific regions in the image (faces, objects)

Basic Section

The basic section contains essential information about the image:

"basic": {
    "title": "Mountain Landscape",
    "description": "AI generated mountain landscape with snow and trees",
    "keywords": ["mountains", "landscape", "snow", "trees", "AI generated"],
    "rating": 4,
    "creator": "John Doe",
    "rights": "Copyright © 2025 John Doe",
    "create_date": "2025-03-15T12:34:56"
}

Field Descriptions

Field Type Description
title String The title or name of the image
description String Detailed description or caption for the image
keywords Array Tags or keywords that describe the image content
rating Integer User-assigned rating (1-5 stars)
creator String Name of the image creator or photographer
rights String Copyright or licensing information
create_date String Creation date in ISO 8601 format

Analysis Section

The analysis section contains image analysis results:

"analysis": {
    "technical": {
        "blur": {
            "score": 0.92,
            "higher_better": true,
            "timestamp": "2025-03-15T12:34:56"
        },
        "noise": {
            "score": 0.08,
            "higher_better": false,
            "timestamp": "2025-03-15T12:34:56"
        },
        "dimensions": {
            "width": 512,
            "height": 512,
            "ratio": 1.0
        },
        "artifacts": {
            "detected": false,
            "score": 0.03
        }
    },
    "aesthetic": {
        "composition": 7.8,
        "color_harmony": 8.2,
        "overall": 7.9,
        "timestamp": "2025-03-15T12:34:56"
    },
    "pyiqa": {
        "niqe": {
            "score": 3.45,
            "higher_better": false,
            "range": [0, 25],
            "timestamp": "2025-03-15T12:34:56"
        },
        "musiq": {
            "score": 78.2,
            "higher_better": true,
            "range": [0, 100],
            "timestamp": "2025-03-15T12:34:56"
        },
        "clipiqa": {
            "score": 0.82,
            "higher_better": true,
            "range": [0, 1],
            "timestamp": "2025-03-15T12:34:56"
        }
    },
    "classification": {
        "style": {
            "value": "photorealistic",
            "confidence": 0.87,
            "timestamp": "2025-03-15T12:34:56"
        },
        "content_type": {
            "value": "landscape",
            "confidence": 0.94,
            "timestamp": "2025-03-15T12:34:56"
        },
        "has_text": {
            "value": false,
            "confidence": 0.98,
            "timestamp": "2025-03-15T12:34:56"
        }
    },
    "eiqa": {
        "color": {
            "dominant_colors": [
                {
                    "name": "forest_green",
                    "hex": "#228B22",
                    "percentage": 0.35
                },
                {
                    "name": "snow",
                    "hex": "#FFFAFA",
                    "percentage": 0.28
                },
                {
                    "name": "sky_blue",
                    "hex": "#87CEEB",
                    "percentage": 0.25
                }
            ],
            "harmony": {
                "type": "complementary",
                "score": 0.85,
                "is_harmonious": true
            },
            "characteristics": {
                "temperature": "cool",
                "contrast": "high",
                "saturation": "medium"
            }
        }
    }
}

Technical Subsection

Field Type Description
blur Object Blur detection metrics
blur.score Number Blur score (higher values mean sharper for most detectors)
blur.higher_better Boolean Whether higher score means better quality
noise Object Noise detection metrics
noise.score Number Noise score (lower usually better)
noise.higher_better Boolean Whether higher score means better quality
dimensions Object Image dimension information
dimensions.width Integer Image width in pixels
dimensions.height Integer Image height in pixels
dimensions.ratio Number Aspect ratio (width/height)

Aesthetic Subsection

Field Type Description
composition Number Composition quality score (0-10)
color_harmony Number Color harmony score (0-10)
overall Number Overall aesthetic score (0-10)

PyIQA Subsection

Contains scores from various Image Quality Assessment (IQA) models:

Field Type Description
niqe Object Natural Image Quality Evaluator scores
musiq Object Multi-scale Image Quality Transformer scores
clipiqa Object CLIP-based Image Quality Assessment scores

Each IQA model entry includes:

  • score: The quality score
  • higher_better: Whether higher scores mean better quality
  • range: The score range as [min, max]
  • timestamp: When the analysis was performed

Classification Subsection

Field Type Description
style Object Detected image style information
content_type Object Detected content type
has_text Object Text detection results

Classification entries include:

  • value: The detected classification
  • confidence: Confidence score (0-1)
  • timestamp: When the classification was performed

EIQA Color Subsection

Field Type Description
dominant_colors Array List of dominant colors in the image
harmony Object Color harmony analysis
characteristics Object General color characteristics

AI Info Section

The ai_info section contains information about AI generation:

"ai_info": {
    "generation": {
        "model": "stable-diffusion-v1-5",
        "prompt": "majestic mountains with snow caps, trees in foreground, clear blue sky",
        "negative_prompt": "ugly, blurry, low quality, deformed",
        "sampler": "euler_a",
        "steps": 30,
        "cfg_scale": 7.5,
        "seed": 1234567890,
        "width": 512,
        "height": 512,
        "batch_size": 1,
        "loras": [
            {
                "name": "landscape_lora",
                "strength": 0.8
            }
        ],
        "timestamp": "2025-03-15T12:34:56"
    },
    "workflow": {
        "nodes": {
            "1": {
                "class_type": "KSampler",
                "inputs": {
                    "seed": 1234567890,
                    "steps": 30,
                    "cfg": 7.5,
                    "sampler_name": "euler_a",
                    "scheduler": "normal",
                    "denoise": 1.0
                }
            },
            "2": {
                "class_type": "CLIPTextEncode",
                "inputs": {
                    "text": "majestic mountains with snow caps, trees in foreground, clear blue sky"
                }
            },
            "3": {
                "class_type": "CLIPTextEncode",
                "inputs": {
                    "text": "ugly, blurry, low quality, deformed"
                },
                "is_negative": true
            }
            // More nodes...
        }
    }
}

Generation Subsection

Field Type Description
model String Base model name/identifier
prompt String Main generation prompt
negative_prompt String Negative prompt for undesired elements
sampler String Sampling algorithm used
steps Integer Number of sampling steps
cfg_scale Number Classifier-free guidance scale
seed Number Generation seed value
width Integer Generated image width
height Integer Generated image height
loras Array LoRA models used in generation

Workflow Subsection

Contains the complete ComfyUI workflow data, including:

  • Node definitions and connections
  • Model parameters and settings
  • Processing pipeline structure

This preserves the entire generation process for reproducibility.

Regions Section

The regions section contains information about specific regions in the image:

"regions": {
    "faces": [
        {
            "type": "Face",
            "name": "Person 1",
            "area": {
                "x": 0.2,
                "y": 0.3,
                "w": 0.1,
                "h": 0.15
            },
            "extensions": {
                "eiqa": {
                    "face_analysis": {
                        "gender": "female",
                        "age": 28,
                        "emotion": "happy",
                        "scores": {
                            "confidence": 0.94,
                            "quality": 0.87
                        }
                    }
                }
            }
        }
    ],
    "areas": [
        {
            "type": "Object",
            "name": "Mountain",
            "area": {
                "x": 0.1,
                "y": 0.2,
                "w": 0.8,
                "h": 0.5
            }
        }
    ],
    "summary": {
        "face_count": 1,
        "detector_type": "deepface"
    }
}

Faces Subsection

Contains detected faces with:

  • type: Face region type
  • name: Face identifier/name
  • area: Position and size (normalized 0-1 coordinates)
  • extensions: Additional analysis data

Areas Subsection

Contains other detected regions with:

  • type: Area/object type
  • name: Area identifier/name
  • area: Position and size (normalized 0-1 coordinates)

Summary Subsection

Contains summary statistics about detected regions.

Additional Metadata Formats

XMP Format

The metadata system complies with XMP (Extensible Metadata Platform) standards:

  • Dublin Core (dc) Namespace: For basic properties (title, description)
  • XMP Basic (xmp) Namespace: For general properties (rating, dates)
  • MWG Regions (mwg-rs) Namespace: For face/area information
  • Custom Namespaces: For AI and analysis data

Exif Format

Embeds metadata in EXIF headers for formats that support it:

  • Title → Exif.Image.ImageDescription
  • Description → Exif.Photo.UserComment
  • Rating → Exif.Image.Rating

IPTC Format

Uses IPTC fields for maximum compatibility:

  • Title → Iptc.Application2.ObjectName
  • Description → Iptc.Application2.Caption
  • Keywords → Iptc.Application2.Keywords

Storage Strategy

The system uses a multi-target storage approach:

  1. Embedded Metadata: Directly inside image files where supported
  2. XMP Sidecar Files: Standard .xmp files alongside images
  3. Text Files: Both machine-readable and human-readable formats
  4. Database Storage: Structured SQLite database for advanced queries

This ensures maximum compatibility and accessibility across different workflows and applications.

Schema Evolution

The metadata schema is designed for easy extension:

  • New Analysis Types: Can be added under the analysis section
  • New AI Models: Supported through the flexible generation structure
  • Custom Fields: Can be added at any level without breaking compatibility
Reference in New Issue View Git Blame Copy Permalink