VeriFile-X runs thirty independent forensic signals — diffusion-reconstruction error, CLIP embedding drift, sensor-noise fingerprinting, compression analysis, and camera metadata — then returns a fully explainable evidence report in seconds. No black-box score. Every signal shows its work.
Built for journalists, legal professionals, researchers, and anyone who needs to verify the authenticity of digital images.
Thirty independent signals — DIRE diffusion-reconstruction error, CLIP embedding distance, PRNU sensor-noise residuals, and a 19-part statistical ensemble. Each signal explains its own reasoning.
Every image is hashed with SHA-256 and MD5, fingerprinted perceptually, and checked for tamper indicators — producing a verifiable evidence trail, not just a verdict.
Analysis runs in memory — your image is never written to disk. A verdict returns in seconds, with a full PDF forensic report available on demand.
Four steps from upload to forensic-grade evidence report.
Drag and drop or select a JPEG, PNG, WebP, TIFF, or HEIC file up to 10 MB.
30 detection signals run in parallel across 8 independent methods.
An AI generation probability score and classification are returned with full signal breakdown.
Export a complete PDF forensic report including hashes, EXIF data, and all signal results.
Drop a file below. A verdict, evidence hashes, and signal breakdown return in seconds.
JPEG · PNG · WebP · TIFF · HEIC · Max 10 MB
DIRE reconstruction, CLIP embeddings, PRNU noise mapping — 30 signals in parallel
Evidence ID: —
Every verdict includes a full per-signal breakdown, not a single opaque score. Not intended as sole evidence in legal proceedings.