A second opinion on every video on your screen.
reads your screen, scans every clip with a ten-metric forensic pass, and gives you a verdict in five seconds. Start free. Go Premium for unlimited scans, real-time alerts, and a 30-day free trial.
Three principles
The shape of a tool you can actually trust to be wrong sometimes.
We picked three things, and held them. Everything else is a consequence.
- 01
Privacy by architecture
We collect the minimum we need. Frames go from your machine straight to Google Gemini. No analytics SDKs, no ad tracking, no cross-app profiles.
Read more - 02
Open methodology
Every metric we score, every weight we apply, every confidence gate is published. Read the math, then decide if the verdict is worth your trust.
Read more - 03
Honest about limits
We have a verdict tier reserved for ‘we are not sure’ and we use it. Real cameras produce false positives. We say so on every page that needs to say so.
Read more
Demo loop
What a verdict actually looks like.
A frame pair, fifty milliseconds apart. Bounded artifacts on the second frame. Score 87. Verdict AI Generated, confidence HIGH. Static composition until the live demo loop ships with v1.0.
F1
F2Capabilities
Built for the people who actually have to verify the video.
VERI-fy was built for a journalist on deadline, not a marketing page. Every capability below is implemented today and visible in the app. Anything we cannot back, we do not list.
Two-stage forensic pipeline
A web-grounded context pass runs only when the model self-flags it. Then a structured 10-metric visual verdict over a frame pair captured 50 ms apart.
Named AI-tool watermark detection
Sora, Veo, Runway, Pika, Kling, Luma, Midjourney, HeyGen, D-ID, Synthesia. A detected watermark settles the verdict.
Bounded artifact regions
Every cited anomaly has a bounding box on the frame. You see what we saw, not just what we concluded.
Stateless per scan
No memory between scans. Each verdict is judged on the frame pair in front of it. No bias from earlier confidence.
Confidence gates with a dead zone
Scores 40–60 are reserved for LOW confidence only. We would rather sit in the middle than be confidently wrong.
Privacy by architecture
No backend, no telemetry, no account. Your Gemini key lives in process memory. Detection history lives in RAM and dies with the app.
Score · 0 to 100
Four bands. The middle one means “we are not sure.”
The 40–60 band is a deliberate dead zone reserved for LOW confidence. When we do not know, we say so. HIGH and MEDIUM confidence cannot land there.
- 0Clean0–25
No anomalies detected. Treat as ordinary footage.
- 26Suspicious26–50
Some signal, no clear evidence. Verify by other means.
- 51Suspect51–75
Concrete artifacts cited. Treat as questionable.
- 76Critical76–100
Strong, multi-metric evidence. Treat as likely AI-generated.
Honest about limits
What VERI-fy is not.
- Not forensic evidence. VERI-fy outputs indicators and scores, not proof. Do not use a verdict in legal, medical, financial, or safety-critical decisions.
- Not perfect. The model produces false positives and false negatives. Real cameras produce compression, color grading, and unusual framing, and sometimes we will flag those. We tell you when our confidence is LOW.
- Not a stream scanner. We sample every 5 seconds, in pairs about 50 ms apart. The HUD updates as fast as the model returns.
- Not your average SaaS. A lightweight account keeps Free Tier fair, and that is the only thing we run a server for. Detection itself stays on your machine, talking directly to Google Gemini. We do not see your screen.
“Tight crop with stylized color grading and slight motion blur. Nothing in the frame pair fires the common-sense check. We cannot cite a specific anomaly. Treat as inconclusive.”
Get VERI-fy
Run a forensic pass on the next video that gives you pause.
Start free with 6 video and 10 photo analyses every month. Try Premium for 30 days, no charge. Windows today, macOS and Linux on the waitlist.
18+ only · KVKK + GDPR compliant · Cancel anytime