AI Image Detector
Estimate whether an image is AI-generated or an authentic photograph using multi-signal heuristics: metadata inspection, C2PA credentials, camera EXIF, and high-frequency noise analysis. 100% private in-browser detection.
Upload image to estimate AI likelihood
Drag & drop JPG, PNG, or WebP photo or click to browse from device
Supports JPG, PNG, WebP, AVIF, TIFF • 100% In-Browser Privacy • Max 50MB
How AI Image Detection Heuristics Work
Understanding signal confidence, compression limits, and authentic verification
Single black-box neural networks often produce high false positive rates when applied to compressed web images. ECSOT cross-references embedded generation parameters, C2PA provenance signatures, optical camera EXIF profiles, and high-frequency edge textures for a balanced estimate.
When an image is shared via platforms like WhatsApp, Twitter/X, or Instagram, ordinary EXIF and camera metadata are automatically stripped for privacy. This causes authentic photos to lack camera tags, correctly yielding an 'Uncertain' score rather than a false definitive claim.
Major generative AI platforms (Adobe Firefly, DALL-E 3, Stable Diffusion WebUI, Midjourney) increasingly embed cryptographically signed C2PA credentials or PNG parameter chunks. When present, these allow 100% verified attribution.
All calculations—decoding, discrete cosine transforms (DCT), perceptual hashing, and metadata inspections—occur exclusively inside your browser's WebAssembly/JS engine. Your sensitive images are never uploaded to any third-party server or cloud database.
How it works?
Upload or Drop Image
Select any JPG, PNG, or WebP photo or synthetic creation from your device.
Multi-Signal Forensic Scan
The client-side engine evaluates embedded metadata, EXIF camera signatures, C2PA claims, and high-frequency noise textures.
Calculates Synthetic Probability
Aggregates signals into a transparent 0%–100% synthetic probability with clear confidence calibration.
Audit Findings & Verification
Review the granular signal breakdown, examine perceptual hashes (pHash), or inspect AI prompts in dedicated tools.