OwnLikeness detects signs of AI-generated voice and likeness use in audio and video files — locally, privately, and under your control — and produces a timestamped evidence receipt you can hand to counsel or your union rep.
If you've heard a clip and thought "that's my voice, but I never said that," you already know the problem. Suspicion isn't evidence. And forensic audio analysis is expensive and slow — the kind of thing you usually need before you know whether you have a case worth pursuing.
OwnLikeness is the first-screen you didn't have. Fast, honest about what it can and can't tell you. The output is a scored evidence receipt — a starting point, not a conclusion — that you can bring to counsel, your union rep, or a platform's trust-and-safety team.
Historical legislative design notes informed this prototype. Current law, legal requirements and suitability of this draft have not been verified. A qualified reviewer must determine the applicable process before use.
Hosted demo at ownlikeness.com: a file you upload here is sent to the Honeycutt Ai Labs server for analysis. It is scored, a receipt is written to a local database on that server, and the audio/video bytes themselves are not retained after scoring beyond the process memory of that request — no separate media archive is kept server-side. The receipt (per-signal scores, hashes, timestamp) persists in the hash-chained case store so you can retrieve it later; the underlying media file does not. This is a demo host operated by one person; treat it accordingly and do not upload anything you would not want processed off your own machine.
Run it yourself: the same code is published in the OwnLikeness README's "Run (local)" section. Anyone who wants nothing to leave their machine at all can run the FastAPI service locally and never send a file to this or any other server. This is the only mode with a zero-network guarantee.
Three signal layers run on every scan. Each reports separately with its own confidence band — you see exactly what fired and why.
Examines the file's own metadata for encoder and software signatures — including known TTS pipelines like ElevenLabs, Bark, Suno, XTTS, Coqui, WaveNet, and voice-clone tools like RVC. Also flags sample-rate anomalies characteristic of desktop synthesis.
Computes an MFCC-hash and spectral consistency statistics — centroid, flatness, bandwidth. These become load-bearing when combined with a performer's registered baseline (Phase 5) for comparison against a suspect file.
A wav2vec2-based deepfake detector runs against the audio. Reports its confidence that the sample is synthesized versus naturally recorded. Explicitly a soft signal — capped in confidence because modern generative TTS regularly outperforms open-source detectors.
A structured, printable, timestamped document. Locally stored. Chained by cryptographic hash so tampering is detectable. Below is a mock — real receipts use your file, not this sample.
Upload a suspect file. Get a scored, timestamped receipt you own and control. On the hosted demo your file is processed on the Honeycutt Ai Labs server and is not retained after scoring; run the published code yourself and nothing leaves your machine (see "Where your file goes"). Attach the receipt to a union complaint or bring it to counsel with something in hand.
Receive a structured, per-signal breakdown from your client — a documented first-screen that can accompany a demand letter or justify commissioning formal expert analysis. Starts the case with a receipt, not a suspicion.
Deploy locally as a first-pass triage layer for performer-flagged content without routing media through a third-party API. Per-signal auditability supports transparency reporting and regulatory inquiry.
We ship in the open so you can see what the tool does today and what's still coming. No overpromising.
"Live" means built, tested, and running at ownlikeness.com today. "Planned" has not shipped yet.
Historical legislative design notes informed this prototype. Current law, legal requirements and suitability of this draft have not been verified. A qualified reviewer must determine the applicable process before use.
As of September 28, 2026 the NO FAKES Act (S.4591 in the Senate, H.R.8915 in the House, 119th Congress) is a bill, not a law. It passed the Senate Judiciary Committee on June 18, 2026 and awaits floor action. OwnLikeness is built to that anticipated framework; it does not claim compliance with it, and the receipts stand on their own under existing state right-of-publicity law.
OwnLikeness is not a legal filing tool. Evidence receipts should be reviewed by qualified counsel before use in any legal proceeding. This tool documents; it does not adjudicate.
Small lab, serious tools. We build detection and epistemic-hygiene software that treats users as capable adults — score with confidence bands, per-signal breakdown, honest disclosure of what a tool can and can't do.
OwnLikeness is one of a sibling family of products built on the same composite-signal-scoring pattern. The pattern is mature. This modality is new.
The hosted demo at ownlikeness.com is live. Access isn't self-serve: email hello@honeycuttailabs.com or use the form below, and we provision your own account by hand — there is no shared or default login. Priority for SAG-AFTRA members, entertainment attorneys, and platform trust-and-safety teams. No spam. No newsletter drip.