I didn’t start out trying to replace Microsoft Azure Face API.
I chose it because it felt like the safe choice: Microsoft, enterprise-grade, compliant, and well-documented. On paper, Azure Face API checks all the boxes for facial recognition.
In practice, using it in a real product pushed me to look elsewhere.
This page is my experience — what worked, what didn’t, and how the main alternatives actually compare when you’re shipping something that matters.
1. Where Azure Face API worked — and where it started slowing me down
Azure Face API does real facial recognition. That’s its biggest advantage over Google’s Vision API.
It supports:
- Face detection with landmarks and pose
- 1:1 face verification
- 1:N face identification
- Person Groups and Large Person Groups
- Strong enterprise compliance
At first, this was enough.
But Azure’s training-based model quickly became friction.
To identify faces, I had to:
- Create Person Groups
- Upload multiple images per person
- Train the model
- Retrain every time something changed
This added:
- State I had to manage
- Delays during updates
- Complexity in user-facing flows
For internal systems, it’s fine.
For SaaS or consumer products, it slowed iteration noticeably.
2. The real breaking points: scaling, pricing, and missing liveness
Three issues pushed me to evaluate alternatives seriously.
Throughput & scaling
Azure enforces strict request limits, especially early on. Bursty traffic gets throttled, and scaling often means tier upgrades or support tickets. That’s risky if your traffic isn’t perfectly predictable.
Pricing in real flows
Pricing is per transaction, but a single “verify user” flow often becomes:
- Detect face (image A)
- Detect face (image B)
- Verify
That’s multiple billable calls per user action. Costs compound quietly.
No native liveness
This was the biggest problem for me.
Azure Face API doesn’t include built-in active liveness. Printed photos and screen replays can pass unless you bolt on extra services or logic. For identity, fraud prevention, or login, that’s a serious gap.
3. FacialProof – the alternative that matched how I expected Azure to work
FacialProof felt like what I thought Azure would be before I understood its internals.
What stood out immediately:
- No training cycles
- Stateless requests
- Built-in liveness & anti-spoofing
- Direct 1:1 and 1:N recognition
- Faster onboarding (API key, not cloud orchestration)
Instead of managing Person Groups and retraining, I could just:
- Send images
- Get results
- Move on
For identity-heavy, user-facing products, this reduced both engineering effort and operational risk.
4. Amazon Rekognition — more scalable than Azure, but heavier
Amazon Rekognition fixes some of Azure’s pain points, but introduces new ones.
What it does better than Azure:
- No training-based Person Groups
- Strong 1:N face search at scale
- Excellent video support
- Handles very large datasets well
What I struggled with:
- IAM complexity
- Multiple AWS services involved (S3, permissions, sometimes Kinesis)
- Liveness handled separately
- Harder to predict total cost across services
Rekognition is powerful, but it’s not simpler than Azure — just different.
5. Face++, MxFace.ai, and Kairos — where each one fits (and doesn’t)
Face++ – fewer steps than Azure, faster iteration
Face++ felt lighter than Azure:
- Direct compare and search APIs
- No explicit training cycles
- Faster iteration
Trade-offs:
- Pricing tiers and QPS limits need attention
- Liveness is usually an add-on, not core
- Still a platform you need to learn
Good option if you want fewer moving parts than Azure.
MxFace.ai – security-focused, but cost-sensitive
MxFace stood out for security:
- Active and passive liveness
- Face verification and identification
- Biometric-focused positioning
But:
- Very limited free usage
- Subscription-heavy pricing
- Face search and storage often billed separately
It works well when security is critical and budget is less constrained.
Kairos – great for testing, risky for core systems
Kairos was the easiest to start with:
- Instant API keys
- Clear concepts
- Optional on-prem deployment
But I wouldn’t rely on it for core identity:
- Limited liveness
- Reliability concerns at scale
- Pricing ramps quickly after free tier
Great for prototypes, risky for production auth.
6. My honest decision framework (what I’d choose today)
I wouldn’t say Azure Face API is “bad”. I’d say it’s misaligned for many modern products.
I’d still use Azure if:
- I was already fully on Azure
- Traffic was predictable
- Compliance mattered more than speed
- The system was internal or controlled
I’d choose an alternative if:
- Identity is user-facing
- Liveness matters
- I expect traffic spikes
- I want fast iteration
- I want fewer moving parts
For most SaaS and consumer apps, API-first facial recognition platforms beat Azure on speed, simplicity, and real-world usability.
That’s why I moved on.
Leave a Reply