Best Liveness Detection SDKs in 2027: On-Premises Options Compared
Shortlisting a liveness detection SDK used to mean picking the vendor with the best demo. In 2027 the harder question is where the check runs: in a vendor’s cloud, on the user’s phone, or on a server you control. That choice affects data residency, cost at volume, latency, and how much you depend on someone else’s uptime.
This guide compares liveness detection SDKs and services by deployment model, with a focus on on-premises and on-device options. It is not a ranking. The right answer depends on your regulator, your volumes, and your stack, so we focus on what each option publicly states and which questions to ask before you commit. Everything below reflects the vendors’ own pages as of November 2026, and capabilities change, so check each one before you decide. For a shorter overview of what liveness detection is, see our 2026 overview of facial liveness detection.
Four Ways Liveness Detection Gets Deployed
- Managed cloud service – you call a hosted API from your app and backend. Nothing to install on a server, but captures travel to the provider’s cloud.
- Vendor-hosted with client SDKs – a mobile or web SDK captures the face, and the vendor’s service makes the decision.
- On-device – the check runs inside your mobile app, so the selfie does not have to leave the phone.
- Self-hosted server – the check runs in a container or service on infrastructure you control, usually for web flows and uploads.
“On-premises” in this guide means the last two: the decision is made on the device or on a server you operate.
What to Compare
- Deployment and data path: where the image is processed and where it is stored.
- Passive and active modes: a single-capture check for low friction, guided actions for step-up assurance, or both.
- Testing evidence: independent PAD results, and exactly which product version they cover.
- Injection attack handling: what the product does when a virtual camera or injected video replaces the real camera feed.
- Platform coverage: iOS, Android, web, Windows, Linux, and cross-platform frameworks.
- Pricing model: per check, per user, or a license without per-check fees.
A Note on iBeta and ISO/IEC 30107-3
ISO/IEC 30107-3 is the standard for testing and reporting presentation attack detection. iBeta, an independent lab, tests products against it and posts confirmation letters on its PAD confirmation letters page. By iBeta’s own description, a letter says the tested solution is compliant with the standard’s testing and reporting requirements for a specific product version and capture configuration, and the testing does not amount to a certification of the vendor’s product. Level 2 uses more demanding attacks than Level 1, such as 3D masks. When a vendor mentions iBeta, ask for the letter and check that the version you will deploy matches the one that was tested.
The Cloud Baseline
Amazon Rekognition Face Liveness
Amazon describes Face Liveness as a fully managed service. The user moves their face into an on-screen oval while a short selfie video is analyzed, and the service returns a confidence score from 0 to 100, a reference frame, and audit frames. It is priced per check, supports web, iOS, and Android, and is offered in a limited set of AWS regions. It suits teams already on AWS who are comfortable with captures being processed in an AWS region.
Azure Face Liveness
Microsoft’s liveness documentation describes a client SDK plus a service that your backend orchestrates through session tokens. It states that the web and mobile solutions conform to iBeta Level 1 and Level 2 ISO/IEC 30107-3 PAD standards, and it spells out a shared-responsibility model in which you secure your backend, sign your app, and check integrity. Review Microsoft’s current Face access policy before you build.
iProov
iProov’s product overview lists SDKs for iOS, Android, web, Flutter, React Native, and .NET MAUI, and two approaches: Dynamic Liveness, which uses its Flashmark technology to illuminate the device screen during capture, and Express Liveness, a faster oval-alignment capture. The overview describes updates being deployed to the cloud and does not describe an on-premises option, so ask the vendor directly if self-hosting is a requirement.
On-Premises and On-Device Options
Innovatrics
Innovatrics reported that its fully on-device passive liveness check passed iBeta Level 2 conformance testing, detecting all 1,500 attempted spoofs in that test. The announcement is from 2021 and covers the version tested then, so confirm what the current release was tested on. The check runs on the user’s device without exchanging data with a server, as part of its Digital Onboarding Toolkit.
Paravision
Paravision’s liveness page describes passive analysis of a single frame, smartphone capture on iOS and Android, server-side liveness on Windows and Linux, and Docker-based packaging. It states iBeta Level 1 and Level 2 compliance. The page does not mention injection attack detection, so that is one to ask about.
FacePlugin
FacePlugin’s Face Liveness Detection SDK supports both passive checks from a single capture and active challenges such as blinking, smiling, or turning the head, with strictness you can tune per risk level. It runs on the device or on your own server, with no calls out to an external service, and ships for Android, iOS, Windows, and Linux and Docker. Licenses are offline with no per-call fees.
On testing, FacePlugin is transparent about what it has and has not got: the passive liveness detection was evaluated internally using the ISO/IEC 30107-3 methodology against printed photos, screen replays, and 3D masks, and a test report is available on request. That is an internal evaluation, not a third-party lab letter. If a third-party letter is a hard procurement requirement for you, say so early.
Which Fits Which Situation
- Already committed to AWS or Azure and fine with cloud processing: the managed services are the shortest path.
- Strict data residency, air-gapped networks, or regulated data: shortlist the on-device and self-hosted options and confirm nothing leaves your network.
- A third-party PAD letter is mandatory: start from iBeta’s list and match the version tested to the version you will run.
- High volumes and cost predictability: compare per-check pricing against a license without per-check fees at your real monthly volume and at ten times that.
- Different risk at different steps: look for both passive and active modes so you can step up only when needed.
Questions to Ask Every Vendor
- Which exact product version and configuration were tested, and by whom?
- Where is each image processed, how long is it kept, and can the whole flow run inside our network?
- How do you handle injection attacks such as virtual cameras and injected video?
- What does pricing look like at ten times our current volume?
- How are models updated, and how do we test an update before it goes live?
- What happens when the service is unreachable or the device is offline?
How to Run a Fair Proof of Concept
- Use the same test set for every vendor: genuine users on a range of phones and lighting, plus your own spoof samples such as printed photos, screen replays, and masks you are entitled to use.
- Measure both errors: genuine users wrongly rejected and attacks wrongly accepted.
- Test on the weakest devices and slowest networks you support.
- Time the full flow, not just the model, and include your retry and manual-review paths.
- Re-run the test after any model or SDK update.
Putting It Together
Liveness is one layer of a verification flow. It sits in front of face matching and works alongside deepfake detection and ID document liveness detection. Teams building onboarding flows can see how the pieces combine on the Customer Onboarding & eKYC page.
Conclusion
There is no single best liveness detection SDK, only the best fit for your deployment model, risk level, and evidence requirements. Decide where the check must run first, then compare testing evidence, injection attack handling, and total cost on your own data. To evaluate FacePlugin against the same criteria, see the Face Liveness Detection SDK and the pricing page.
Advanced Biometric Solutions at FacePlugin
Transform the way you secure and verify identities with FacePlugin’s on-premises face recognition, face liveness detection, deepfake detection, and ID document recognition SDKs – engineered to keep identity verification secure, private, and fully under your control.
Talk to Our TeamFrequently Asked Questions
What is the difference between cloud and on-premises liveness detection?
With a cloud service, the captured face video or image is sent to the provider's servers for analysis. With on-premises options, the decision is made on the user's device or on a server you operate, so captures can stay on infrastructure you control. The trade-off is that you take on more of the deployment and monitoring yourself.
Do I need iBeta Level 2 testing for a liveness detection SDK?
It depends on your regulator, your risk level, and your procurement rules. Level 2 uses more demanding attacks than Level 1, such as 3D masks, so it is the stronger signal. If you need it, ask for the confirmation letter and check that the product version and configuration tested match what you will deploy.
Should I choose passive or active liveness detection?
Passive liveness checks a single capture and adds little friction, which suits most sign-ups and logins. Active liveness asks the user to blink, smile, or turn their head, which adds assurance for higher-risk steps. An SDK that supports both lets you step up only when needed.
Is a per-check price or a license better for liveness detection?
Per-check pricing is simple to start with but grows with every verification. A license without per-check fees keeps the cost flat as volume rises, but you run the software yourself. Compare both at your real monthly volume and at ten times that before deciding.
Deploy identity verification AI on your own terms
Talk to our team about on-premises face recognition, liveness, deepfake, and ID document verification — or dive into the documentation.