Changelog
Track all updates, new features, and improvements to our AI detection technology
AI Image Detector V2 — LLM-Powered Analysis
Major release introducing AI Image Detector V2 powered by PhotoCheck AI LLM. The homepage now defaults to V2 for all uploads, providing deep AI-generated content analysis with human-readable reports in 30+ languages. V1 Classic (17 methods) remains available at /ai-image-detector-v1.
LLM Detection Engine
- • PhotoCheck AI LLM — Deep visual analysis using Gemini 2.5 Pro, fine-tuned on billions of images for AI-generated content identification
- • SynthID Watermark Detection — Google's invisible watermark detection for AI-generated images
- • Technical Forensic Scanning — Local forensic analysis including metadata, JFIF, noise patterns, and C2PA
- • Human-Readable Reports — Detailed analysis reports with confidence scores and reasoning explanations
- • 30+ Output Languages — Results available in English, Vietnamese, Chinese, Japanese, Korean, and 25+ more
Homepage & UI Changes
- • V2 Default — Homepage upload now uses AI Image Detector V2 (LLM) by default
- • Violet Theme — New violet accent design for V2 pages, distinguishing from V1 green theme
- • 3-Tool Showcase — Feature section displays V2 LLM, Deepfake Detector, and V1 Classic
- • V1 Classic Route — Original 17-method detector accessible at /ai-image-detector-v1
Forensic Analysis Tool Pages
5 new dedicated online tool pages for individual forensic detection methods. Each tool provides a focused landing page with upload capability, SEO optimization, and detailed method explanations.
New Tool Pages
- • Micro Texture Analyzer - Detect AI texture repetition, uniformity, and synthetic randomness patterns
- • Semantic Inconsistency Detector - Find physics violations, shadow errors, and perspective mistakes
- • PRNU Noise Analyzer - Verify camera sensor fingerprint and noise patterns
- • Compression Artifact Analyzer - Analyze JPEG quality, blocking artifacts, and recompression
- • Frequency Spectrum Analyzer - DCT spectrum and Fourier transform analysis
C2PA AI Detection Override & Performance
Smart C2PA content credential override — when an image is certified as AI-generated via C2PA (e.g., by Google, Adobe, OpenAI), the detection score is automatically overridden to 100% with high confidence. Plus performance improvements for faster cold starts.
C2PA Override Logic
- • 100% Override - When C2PA digitalSourceType confirms
trainedAlgorithmicMediaorcompositeWithTrainedAlgorithmicMedia, AI score overridden to 100% - • Deep Manifest Search - Searches entire C2PA JSON including ingredient manifests for AI source types
- • Client-Side Fallback - Frontend also scans raw manifest data for AI types even if backend doesn't extract them
- • Visual Override - Verdict badge, confidence, reasoning, and probability ring all update to show C2PA-verified AI content
- • Human-Readable Annotations - Digital source types shown with plain-language explanations
Performance Improvements
- • Offline Model Loading - AI detection models loaded from local cache without HuggingFace API calls after first download
- • Faster Worker Startup - Eliminated redundant HTTP requests to HuggingFace on every Celery worker restart
- • C2PA CLI Tool - Switched from unstable c2pa-python FFI to reliable c2patool CLI for manifest parsing
Bug Fixes
- • Fixed texture analyzer overflow/divide-by-zero warnings
- • Fixed digitalSourceType extraction missing ingredient manifests
- • Fixed C2PA checker page not showing source type annotations
Advanced Image Tools: Border & Blur
Two new powerful client-side tools with advanced features: Image Border Online and Blur Image Online.
Image Border Online
- • 4 Border Types - Solid, Gradient (linear/radial), Pattern, Polaroid
- • Individual Side Control - Different sizes for top/right/bottom/left
- • Rounded Corners - 0-100% radius slider
- • Drop Shadow - Offset, blur, color customization
- • 6 Quick Presets - None, Thin, Thick, Polaroid, Shadow, Rounded
Blur Image Online
- • 7 Blur Types - Gaussian, Box, Motion, Radial, Pixelate, Lens/Bokeh, Tilt-Shift
- • 4 Selection Tools - Brush, Rectangle, Circle, Lasso
- • AI Face Detection - Auto-detect and blur faces
- • Background Blur - Portrait mode with AI subject detection
- • 6 Quick Presets - Face Privacy, Portrait, License Plate, Text Censor, Tilt-Shift, Bokeh
Technical Details
- • 100% client-side processing - no image uploads
- • HTML5 Canvas API for real-time rendering
- • Export to PNG, JPEG, WebP with quality control
- • Undo/Redo with keyboard shortcuts
Multi-Method Deepfake Detection
Advanced deepfake detection using 4 neural network methods with ELA and Heatmap visualization.
New Features
- • CNNDetection (ResNet-50) - Trained on 720K+ images from ProGAN
- • UniversalFakeDetect (ResNet-50) - Hybrid Swin+EfficientNet architecture
- • ELA Analysis - Error Level Analysis with visual heatmap
- • Heatmap Analysis - Neural activation visualization
- • R2 Storage - Analysis images stored in cloud for faster loading
- • 20 FAQs - Comprehensive FAQ section on deepfake detection
Technical Details
- • Ensemble method combining 4 detection approaches
- • Celery Chord orchestration for parallel processing
- • 93-95% accuracy across DALL-E, Midjourney, Stable Diffusion
- • Dedicated /deepfake-detector page with real-time analysis
Multi-Language Support & Blog
Full internationalization for 30 languages and new blog with AI detection guides.
New Features
- • 30 Languages - All tools available in English, Chinese, Spanish, French, and 26 more
- • Blog Section - 16 articles on AI detection, deepfakes, and image verification
- • Better Navigation - Language preference preserved across all pages
Invisible Watermark Detection
New dedicated tool to detect invisible AI watermarks embedded by Google, Stability AI, and protection tools.
New Features
- • SynthID Detection - Detect Google AI watermarks from Imagen, Gemini, Nano Banana
- • Stable Signature Detection - Identify Stability AI watermarks from Stable Diffusion
- • GLAZE/Nightshade Detection - Find adversarial protection patterns
- • Steganography Detection - Detect LSB hidden data patterns
- • Dedicated Tool Page - /watermark-detector with detailed analysis
- • Ensemble Integration - Results shown in main analysis page
Technical Details
- • DCT entropy analysis for SynthID patterns
- • Channel correlation analysis for Stable Signature
- • Adversarial perturbation detection for GLAZE
- • LSB analysis with Chi-square and RS steganalysis
- • Weighted 8% contribution to overall AI score
17-Method AI Detection System
Major upgrade from 8 to 17 detection methods, significantly improving detection accuracy to 90%+.
New Detection Methods
- • Semantic Inconsistency Detection - Detects logic errors like incorrect shadows and impossible perspectives (11%)
- • Human Biometric Analysis - Analyzes fingers, eyes, and skin texture using MediaPipe (11%)
- • Lighting Physics Validation - Validates light source consistency and shadow physics (9%)
- • Compression Artifact Analysis - Detects JPEG re-compression patterns (7%)
- • Edge Sharpness Analysis - Analyzes depth-of-field consistency (6%)
- • Statistical Pattern Analysis - Uses entropy and Benford's Law (4%)
- • Chromatic Aberration Analysis - Detects missing lens artifacts (3%)
- • Micro-Texture Analysis - Finds texture repetition patterns (3%)
- • Color Palette Analysis - Analyzes saturation and color diversity (3%)
Improvements
- • Overall detection accuracy improved to 90%+
- • Updated ensemble analyzer with rebalanced weights
- • Added 9 new technology documentation pages
- • Improved Schema.org structured data for SEO
Multi-Language Support Expansion
New Languages
- • Added support for 30 languages total
- • Full translation of all technology pages
- • Improved SEO with hreflang tags
- • Language-specific sitemap generation
Image Tools Suite
New Tools Added
- • Image Cropper with aspect ratio presets
- • Image Rotator with flip options
- • Image Filters (grayscale, sepia, blur, etc.)
- • Watermark Tool with text/image support
- • Brightness & Contrast adjustment
- • Collage Maker with multiple layouts
- • QR Code Generator
- • Image to Base64 converter
C2PA Verification
Content Provenance
- • Full C2PA manifest verification
- • Support for Adobe, Microsoft, and OpenAI signatures
- • Provenance chain visualization
- • Dedicated C2PA Checker tool page
GAN Fingerprint Detection
GAN-Specific Analysis
- • Checkerboard pattern detection via FFT
- • Color banding analysis in gradients
- • Spectral anomaly detection
- • Upsampling artifact recognition
Advanced Forensic Methods
New Detection Capabilities
- • PRNU (Photo Response Non-Uniformity) sensor fingerprint analysis
- • Gradient analysis for edge patterns
- • Texture analysis using Local Binary Patterns
- • Noise pattern uniformity detection
Initial Release
First public release of AI Photo Check - the free AI image detection tool.
Core Features
- • ML-based detection using Vision Transformer models
- • Frequency analysis with DCT spectrum
- • Metadata analysis for EXIF and AI tool signatures
- • User dashboard with analysis history
- • Google OAuth authentication
- • Responsive design with dark theme
Try Our Latest Detection Technology
Experience the power of 17 advanced AI detection methods
Попробовать