Synthetic-media detection
Almost every deepfake detector is trained to recognise known fakes — so the moment a new generator ships, it goes blind. We flip the problem: we model what REAL footage looks like and flag the departure. A generator we never trained on is caught by construction — and we show you exactly WHERE in the frame the video stops behaving like reality.
Why every "98% accurate" claim collapses in the wild
A detector that learns the tells of yesterday's generators is playing a game the defender is guaranteed to lose — the attacker always has the newer model. Independent research is blunt about it: state-of-the-art detectors lose roughly half their accuracy moving from academic test sets to real-world clips, and some fall to a coin flip.
The diagnosis, in the researchers' own words: a model trained on specific generators learned those generators' specific traces, not a general theory of forgery.
Senua never studies any particular generator. It builds a rich model of what genuine capture looks like, then flags any video that departs from it. That is the entire moat: generator-agnostic by design, and therefore future-proof against the next model instead of one step behind it.
How it works
Instead of learning what fakes look like, Senua learns what real looks like — the way genuine footage is produced, the way real motion and light and physiology behave — and flags what departs from it.
Layer 1
Genuine footage is created and encoded by a real camera-and-processing pipeline, which leaves consistent, hard-to-fake traces in the file and stream itself. Synthesis and re-encoding pipelines leave different ones. Senua reads this fast, as a first pass — before looking at a single pixel of content.
Layer 2
Real video is coherent frame to frame in ways generators still struggle to sustain — motion obeys physics, light and shadow stay consistent as a scene moves, and a real face carries faint involuntary rhythms (the colour pulse of blood flow, natural blink and micro-expression timing). Senua models these authentic dynamics and flags footage that stops behaving like the real thing.
Layer 3
Most fakes don't replace the whole picture — a face is swapped onto a real body, an object inserted into real footage, one region quietly regenerated. Senua checks every region of every frame independently and paints a heatmap over the video highlighting exactly the parts that stop behaving like reality, while the authentic background stays clear.
It doesn't just say "fake"
The reviewer doesn't get a number to trust — they get to see the manipulation, where it is and when it happens. Senua checks every region of every frame independently and highlights exactly the parts of the picture that stop behaving like reality, frame by frame, while the authentic background stays clear.
Not "fake, 0.8" — you watch the swapped face light up. This is the highest form of explainability in the market, and it is exactly what convinces a skeptic, closes a fraud case, or stands up in front of a jury.
What a result looks like
Every verdict names WHAT departed from authentic — which region, which frames, which signal — with a confidence attached. Confidence is reported honestly, including "uncertain."
A partial face-swap: the face region fails the authenticity signature while the room around it holds.
A heavily-compressed, very short clip: too little signal to decide. Senua says so rather than guessing.
No detector is infallible, and we will never claim otherwise — a claim of perfection is itself a red flag. What Senua offers is the thing the market is missing: detection that does not have to have seen the generator first, that runs where the video actually is, with an explanation you can defend.
This public tool is a hosted demo of an engine that ships on-device. For the demo, your clip is uploaded, analyzed, and deleted immediately — nothing is retained unless you explicitly opt in to help improve detection. The enterprise product runs entirely on your own hardware; your video never leaves your machine.
Get access → Request a link, paste a clip from this week's newest generator, and watch it get caught.