Fast — mixed-10-v2¶

The fastest model. Analyzes 10 frames (0.3 seconds of video) for the lowest possible latency.

At a Glance¶

Model ID	`mixed-10-v2`
Frames required	10
Capture time	~0.3 seconds at 30 FPS
Avg response time	~1s
Security level	Strong

When to Use¶

High-volume onboarding where drop-off from wait time is a concern
Repeat verification (e.g., re-auth for returning users who are already trusted)
Mobile-first flows where users expect instant results
Low-risk transactions where speed matters more than maximum security

Performance¶

Based on evaluation across 1,960 test videos (9 attack types, held out from training):

EER	4.4%
AUC	0.988
Balanced accuracy	95.2%
Avg response (fast-check)	~1s
Avg response (fast-check-crops)	~1s

Per-Attack Detection¶

Attack Type	Detection Rate
Physical spoofing	97.7%
Deepfakes	93.3%
Screen replay	92.6%
Faceswap	85.2%

What this means

At the EER operating point, 95.6% of attacks are correctly blocked and 95.6% of legitimate users pass. With 10 frames of capture, this is the fastest path to a strong liveness decision.

Tradeoffs¶

Fastest response — minimal wait time for users
Strong across all attack types — 93%+ detection for deepfakes, physical spoofing, and screen replay
Faceswap is the hardest category — 85% detection; consider escalating borderline results to Balanced

Integration ¶

{
  "session_id": "550e8400-e29b-41d4-a716-446655440000",
  "model": "mixed-10-v2",
  "source": "live",
  "frames": [
    { "index": 0, "timestamp_ms": 0, "pixels": "..." },
    { "index": 1, "timestamp_ms": 100, "pixels": "..." },
    ...
  ]
}

Capture 10 frames at 30 FPS (0.3 seconds total).