Verify Any Video, Image, Audio Clip, or Document for AI Generation — With Verifiable Proof.

Cryptographic & container provenance: C2PA Content Credentials are cryptographically signed manifests added by tools like Adobe Firefly, OpenAI, and Leica/Sony cameras. When valid, they provide cryptographic proof of origin or edit history.

What happens on social media: Platforms like WhatsApp, X, and Instagram routinely strip both C2PA and camera EXIF headers. When metadata is stripped, TruthLens reports the metadata absence honestly and gives you a 1-Click OSINT Reverse Image & Keyframe Launchpad (Google Lens, TinEye, Bing, Yandex) plus SHA-256 and perceptual dHash fingerprints to trace the earliest online appearance.

TruthLens layers multiple independent checks rather than relying on a single black-box number:

1. Verifiable Provenance: C2PA signatures, XMP/IPTC digital-source tags, embedded AI generation parameters, and invisible watermark carriers.
2. Multi-Model Neural Triage: Ensemble vision classifiers for images and 4-second sliding-window Wav2Vec2 classifiers for audio.
3. Spatial & Spectral Signals: Error Level Analysis (ELA), 2D Fourier frequency spectra (FFT), 3×3 regional noise consistency grids, and STFT audio spectrograms.

When signals are weak or contradictory, TruthLens returns Inconclusive rather than guessing.

When the generator saved workflow metadata in the file, yes. TruthLens extracts prompts, negative prompts, checkpoints, LoRA modules, sampler settings, seeds, and img2img denoise strength from PNG text chunks and JPEG/WebP EXIF UserComment fields written by AUTOMATIC1111, ComfyUI, SwarmUI, Fooocus, InvokeAI, and NovelAI. We never hallucinate prompts from pixels—if the metadata was stripped, we tell you clearly.

TruthLens combines writing-style & neural signal measurements with document container provenance:

• Style & Model Signals: Measures formulaic stock phrases, sentence length variation (burstiness), vocabulary variety (MATTR), DistilGPT-2 token perplexity, and sliding-window neural text classification.
• Document Provenance: For uploaded .docx and .pdf files, TruthLens reads OpenXML revision counts, logged editing minutes, authoring software, and PDF incremental save markers (%%EOF).

These signals describe writing style and file save history—not definitive proof of who wrote the text.

It highlights stock phrases and overly long, uniform sentences and offers plainer, more direct rewrites. Applying a rewrite edits your draft in the workbench; click Re-scan to recompute the style and perplexity measurements.

The coach is designed to help writers produce clearer, more specific prose—not as an adversarial "stealth bypasser."

Not stored: Uploads and pasted text are processed in volatile memory to produce your analysis response and are not persisted to disk or a database. PDF reports are generated locally in your browser.

No training: We never use your content to train or fine-tune AI models. See our Privacy Policy for full details.