How it Works Applications Roadmap
 BOOK A DEMO
● MoodMetrics AI has been accepted into the MassChallenge 2026 Cohort
// Emotion-Native Verification
We don't detect the fake.
We verify the real.
Our product verifies people by the one thing generative models can't reproduce: the way they genuinely feel. We read emotional behaviour across face, voice and language, and we're moving it from forensic to real time.
Built & Backed By
// Methodology
The structural gap
There's a widening gap between yesterday's artefact-hunting detectors and today's generative fraud. Closing it means rethinking what we verify in the first place — not the pixels, the person.
YESTERDAY'S APPROACH
Pixel-based detection
Scrutinising the container for technical imperfections and artefacts.
ARTEFACT HUNTING
Scans for compression errors and rendering glitches, which generators are already learning to eliminate.
STATIC FRAME ANALYSIS
Judges a single frozen image, blind to the temporal sequence where deception actually lives.
ARMS-RACE DEPENDENCY
Accuracy degrades with every new model release, needing constant and costly retraining just to stay relevant.
THE NEW STANDARD
Multimodal emotion verification
Verifying the source by reading the involuntary emotional signals no synthetic model can fake.
BEHAVIOURAL SIGNAL CAPTURE
Reads how a specific person modulates emotion across face, voice, and language, to create a signature trained on them, not the general population.
CROSS-MODAL COHERENCE
Authentic humans shift face, voice, and words together. Deepfakes synthesise each channel alone and break the alignment.
MODEL-AGNOSTIC BY DESIGN
Anchored to human physiology, not generator architecture. It gets stronger as attacks evolve, not weaker.
// HOW IT WORKS
Three signals. One person.
Impossible to perfectly replicate.
Our service builds an emotional baseline for each person you want to verify videos for, then measures how far a new video drifts from it, across every modality at once.
FACE
Per-frame scores from facial emotion expressions, e.g. contempt, embarrassment, amusement etc. The involuntary tells.
PROSODY
Emotion derived from vocal pitch, energy, rhythm and pace. How excitement and stress move the voice is deeply individual.
LANGUAGE
Semantic emotional valence from what's actually said, the third channel that has to stay coherent with the other two.
We analyse 148 emotion signals every second, across any demographic, in 42 languages.
AUDIO-ONLY MODE SUPPORTED
If there is no face in the video, we can verify using prosody and language alone. All three modalities are used when available.
THE PIPELINE · RAW VIDEO → AUTHENTICITY CONFIDENCE VALUE
01
Ingest
Raw video in. No frames written to disk, fully in-memory and stateless.
02
Extract
Extract the authentic frame and utterance-level emotion signals across face, prosody and language.
03
Transform
Compress those signals into a fingerprint of the person's authentic emotional pattern.
04
Compare
Measure how closely new video matches the enrolled authentic baseline.
05
Score
A single confidence value between 0.0 and 1.0, ready to act on.
// THE OUTPUT
One number you can act on.
Every verification returns a single authenticity confidence value, a calibrated similarity score, not an opaque verdict. Wire it straight into your risk thresholds.
SYNTHETIC
AUTHENTIC
0.94
// APPLICATIONS
Wherever you have to ensure that a face or voice is genuine.
01 · ENTERPRISE SECURITY
Executive & customer impersonation
AI voice and video cloning targeting your C-suite is now a mainstream attack vector. We verify that the person on the call is who they claim to be, before a wire transfer is authorised or a decision is made.
02 · COMMUNICATIONS INFRASTRUCTURE
Secure video conferencing
Our API integrates directly into your existing meeting infrastructure via a lightweight bot or desktop agent, flagging synthetic participants in Zoom, Teams, and Meet without requiring any changes to the platform itself.
03 · FINANCIAL SERVICES
Call centre verification
Voice deepfakes are the fastest-growing fraud vector in financial services. We stop them before the damage is done.
// FOR DEVELOPERS
A few requests away from a verdict.
REST over HTTPS, API-key auth, async job polling.
Upload a video, poll the job, read the results.
POST
/subjects
Create a subject
POST
/subjects/{subject_id}/videos
Add videos to fingerprint
POST
/subjects/{subject_id}/verify
Verify videos against fingerprint
GET
/jobs/{job_id}
Retrieve job details
curl -X 'POST' \ 'https://mithril.moodmetrics.ai/api/subjects/$SUBJECT_ID/verify' \ -H 'accept: application/json' \ -H 'Authorization: Bearer $API_KEY' \ -H 'Content-Type: multipart/form-data' \ -F 'videos=@SUBJECT_VIDEO.mp4;type=video/mp4'
// THE ROADMAP
From forensic to real time.
● LIVE NOW
Async verification API
Submit recorded video, receive a calibrated authenticity confidence score. Live and ready to integrate.
IN BUILD
Integrated pipeline
Fully managed, auto-scaling cloud deployment. Verification embedded directly in your organisation's communication stack: Teams, Meet, Slack, WhatsApp, and beyond.
THE VISION
Real-time verification
Authenticity scored live, mid-call. Deepfakes are flagged the moment they appear.
// REQUEST ACCESS
Bring us your hardest deepfake.
Book a 30-minute call. We'll run a live verification on a sample of your choosing and show you the score, and how to wire it into your stack.
BOOK A DEMO →

Or contact us at

kyle@moodmetrics.ai