What do you want to verify?
Each modality has its own forensic playbook. Pick one to begin.
Default to on-device when we can. Image, text, and C2PA verification run client-side via ONNX Runtime Web / WASM — the file never leaves your machine. Heavier diffusion-reconstruction and video models escalate to the cloud only when you opt in. Bundle budget: ~1–12 MB per modality, lazy-loaded.
Text
liveBurstiness, lexical tics, n-gram repetition, sentence-start variety, punctuation patterns.
Open detector →
Image
liveError Level Analysis, FFT spectrum, channel decomposition, noise residual, C2PA provenance.
Open detector →
Audio
liveSTFT spectrogram, spectral flatness uniformity, harmonic stability, centroid drift. mp3, wav, m4a, ogg, flac.
Open detector →
Video
live12-frame sampling, noise consistency, lighting plausibility, frame-to-frame motion, key-frame ELA + noise residual.
Open detector →