Identity Verification API
⚡ Performance KPIs (measured)
| Metric | Brainiall | AWS Rekognition | |
|---|---|---|---|
| End-to-end pipeline p50 latency | 154 ms [source] | 250–500 ms [source] | ★ |
| Face detection p50 (component) | 8 ms [source] | ~50–100 ms (cloud RTT incl.) | ★ |
| Face embedding p50 (component) | 59 ms | ~80–150 ms | ★ |
| Anti-spoof p50 (component) | 87 ms | Liveness ~200 ms (multi-frame) | ★ |
| Throughput per CPU core (sustained) | 8 RPS | Cloud-managed (auto-scale) | = |
| Cold-start | ~6 s (always-warm container) | <1 s (warm pool) | R |
🎯 Capability matrix
| Metric | Brainiall | AWS Rekognition | |
|---|---|---|---|
| Face Detection | ✅ Brainiall face detector | ✅ DetectFaces | = |
| Face Recognition / Compare 1:1 | ✅ Brainiall Identity engine v1 (512-d embedding) | ✅ CompareFaces | = |
| Face Search 1:N | ✅ via embedding + cosine threshold | ✅ SearchFacesByImage | = |
| Anti-Spoof / Liveness | ✅ Brainiall deepfake module | ✅ Face Liveness (iBeta certified) | R |
| iBeta PAD formal certification | ❌ Not yet (planned 2026) | ✅ Level 1 + 2 | R |
| LGPD / GDPR-by-default residency | ✅ EU/BR datacenter (Latitude FRA/MAD) | 🟡 us-east default; EU regions extra | ★ |
| Local CPU deployment (no cloud RTT) | ✅ Latitude bare-metal | ❌ Cloud-only API | ★ |
| Open weights you can audit | ✅ permissive licenses (Brainiall face detector, Brainiall Identity engine, Brainiall deepfake module) | ❌ Proprietary closed | ★ |
📊 Quality benchmarks (literature)
| Metric | Brainiall | AWS Rekognition | |
|---|---|---|---|
| Face detect benchmark | Brainiall face detector ~95% WIDER Easy / 88% Hard | Not published (claim '99.9% confidence') [source] | |
| Face recognition LFW pairs | Brainiall Identity engine v1 99.4% | Not published (claim '99%+ internal') | |
| Anti-spoof (CelebA-Spoof) | Brainiall deepfake module ~97% acc | iBeta PAD Level 1 + 2 cert [source] | R |
Pricing
Free
10 KYC checks/month
Try the API end-to-end. No card required. KYC carries higher abuse-risk than other SKUs, so the free quota is intentionally smaller.
Fast
$0.005 / check
Brainiall face detector + Brainiall Identity engine + Brainiall deepfake module. p50 154 ms. 8 RPS sustained per worker.
Pro
$0.01 / check
Adds liveness multi-frame, JSON consent log, audit-trail webhooks.
Quickstart (Python)
Request
import base64, httpx
img = base64.b64encode(open("photo.jpg", "rb").read()).decode()
resp = httpx.post(
"https://api.brainiall.com/v1/identity/verify/base64",
headers={"Authorization": "Bearer brnl-..."},
json={"image": img, "include": ["detection", "embedding", "antispoof"]},
)
print(resp.json())Example response
{
"request_id": "req_8f2a1c…",
"processing_ms": 154,
"faces": [
{
"bbox": [342, 198, 587, 511],
"detection_confidence": 0.998,
"antispoof_label": "real",
"antispoof_score": 0.96,
"embedding": [0.012, -0.043, …], // 512-d vector
"landmarks": {
"left_eye": [398, 286], "right_eye": [519, 280],
"nose": [462, 350], "mouth": [462, 442]
}
}
],
"warnings": []
}Comparison methodology & disclaimer
Brainiall measurements: latency and throughput were measured on Brainiall production infrastructure (Latitude bare-metal · production hardware · 48 vCPU · CPU inference) in May 2026. p50/p95 from 5–6 sample images per model. Models tested: Brainiall face detector (MediaPipe Tasks API, permissive license), Brainiall Identity engine v1 (fal.ai, permissive license), Brainiall deepfake module (prithivMLmods, permissive license). Full report: Phase 1.5 Eval Report.
AWS Rekognition data: derived from publicly available AWS documentation, official blog posts, and AWS service cards as of May 2026. AWS does not publish formal benchmarks on standard datasets (LFW, WIDER FACE, MegaFace, IJB-C); their published claims (e.g. "99.9% confidence", "iBeta PAD Level 1+2") are cited verbatim with source links per row. Latency ranges for Rekognition reflect public AWS docs and customer reports including network round-trip from us-east-1; your latency may vary by region and by image size.
Important notes:
- Methodologies and evaluation datasets may differ between Brainiall and AWS — direct head-to-head benchmarks on identical input are scheduled for Q3 2026 and will be published as a separate report.
- Quality benchmarks for our models (LFW, WIDER FACE) come from the original model papers — independent reproduction may yield slightly different numbers.
- Pricing comparison: AWS Rekognition charges per API call (e.g. $0.001/face detect, $0.0025/liveness check); Brainiall charges per pipeline check at the indicated tier.
- iBeta PAD certification: AWS Rekognition has formal iBeta level 1 and 2 cert; Brainiall's anti-spoof is based on Brainiall deepfake module which has not undergone iBeta evaluation. For regulated KYC workloads requiring iBeta, AWS Rekognition currently has a clear edge.
- Trademarks: Amazon Web Services and Rekognition are trademarks of Amazon.com, Inc. or its affiliates. This page is provided for informational comparison purposes and is not endorsed by or affiliated with AWS.
Last reviewed: May 2026. Source citations are attached per metric where applicable. We update this page when new benchmark data becomes available.