Release 8.3 — August 4, 2026

Important Changes

Added support for a new liveness type: Passive Liveness with Blink. To complete the check and confirm liveness, the user should blink at the end of the process.
Updated the neural networks used for liveness presentation attack detection to improve success rates for real users and enhance protection against fraudulent attempts.
The Face SDK is now modular. You can load only the components required for your case. For example, deployments that use only Face Comparison no longer need to load Liveness components into memory. This reduces memory consumption and can lower deployment costs by allowing more workers to run and process more requests in parallel. For configuration details, see Initialization Policy.
Optimized Match and Detect requests. In internal tests on NVIDIA RTX 5050 with two workers, performance improved by up to 2.5x.
Liveness checks now require at least HD-quality frames. In version 8.3, this requirement is validated only by the Mobile SDK and Web Components. Mobile SDKs and Web Components of earlier versions continue to work as before. In future versions, this quality check will also be enabled on the backend. As a result, sessions with low-resolution frames will be rejected regardless of the Mobile SDK or Web Component version used.
Added a new tab to the Demo Website to demonstrate the age estimation capability. As with other tabs, it includes request and response examples. You can try it on the Demo Website and learn more in the Age Estimation documentation.

Updated the neural networks used in liveness checks to detect occlusions and reduce the false positive detection rate.
Added a new prompt to guide users when face occlusion is detected.
Added on-device selfie quality validation during liveness checks. The validation is performed directly on the capture device. If the selfie quality is not sufficient, the user receives an immediate prompt with guidance on what needs to be corrected. The previous lighting condition detector has been deprecated and disabled due to a high false rejection rate.

Added the ability to use livenessTransactionId as a request parameter instead of downloading a selfie and including it in the request. For details, see Regula Face SDK Web API.

Added support for Central Kurdish (Sorani) and Tajik. For the complete list of supported languages, see Face SDK Localization.

Fixed an issue where the first custom image for a liveness prompt was stretched to the full screen.
Fixed an issue where the Android SDK could crash with NullPointerException in some cases when no face was present in the frame.
Fixed an issue where a custom image for the Close button was shaded black.

Added support for a mode that restricts screenshots and screen recordings within the application. For details, see Preventing Screen Capture.
Added support for Mutual TLS. For details, see the Mutual TLS Guide.
Increased the size of the liveness guide circle when using the back camera.
Fixed an issue that caused a crash in RGLCISSyscalls.c.

The /healthz endpoint now retrieves up-to-date license information every time it is invoked. Restarting the service is no longer required to update license information after the license has been renewed or extended.
Added support for graceful shutdown. When Web Service receives a shutdown command, it stops processing incoming requests, completes requests that are already in progress, and closes all storage and database connections.
Improved the diagnostic message displayed when the transaction limit per tag is exceeded.
Improved the stability of the housekeeper when processing large volumes of data. The housekeeper now cleans up data in chunks.
Added support for storing Milvus credentials securely in environment variables. For details, see Order of Precedence.
Fixed an issue where a liveness video could not be saved to Azure Storage and the following error was returned: The metadata specified is invalid. It has characters that are not permitted.
Fixed an issue where it was possible to create a person not linked to any group if an invalid group ID was provided.
Updated dependencies to the latest versions to address security vulnerabilities.

Updated the Demo Website design to make it visually consistent with the Document Reader SDK Demo Website.
Fixed an issue where the Demo Website could become unresponsive during loading.
Fixed an issue where face detection rectangles could be displayed incorrectly if the uploaded image contained an Orientation tag.
Updated dependencies to the latest available versions to address security vulnerabilities.
Known security issue. The PostCSS library remains affected by security vulnerabilities tracked as CVE-2026-45623 and GHSA-r28c-9q8g-f849. We will address these vulnerabilities in a patch release when a fixed PostCSS version becomes available.

Added support for Armenian, Central Kurdish (Sorani), Kyrgyz, and Tajik. For the complete list of supported languages, see Settings.
Updated dependencies to the latest versions to address security vulnerabilities.

Updated the documentation for the Azure deployment package.
Fixed an issue where the Windows installer did not preserve application settings during silent installation when Keep application settings enabled was selected.
Updated the MinIO version in the Docker Compose configuration.

The following Face SDK component versions comply with this release:

Component	Version
Web Service	8.3.749.1905
Web Components	8.3.2309
Mobile Native
ㅤㅤㅤiOS API	8.3.4725
ㅤㅤㅤiOS Core Basic	8.3.2543
ㅤㅤㅤiOS Core Match	8.3.2545
ㅤㅤㅤAndroid API	8.3.5199
ㅤㅤㅤAndroid Core Basic	8.3.1975
ㅤㅤㅤAndroid Core Match	8.3.1975
Mobile Hybrid
ㅤㅤㅤNET MAUI iOS API	8.3.947
ㅤㅤㅤNET MAUI iOS Core Basic	8.3.666
ㅤㅤㅤNET MAUI iOS Core Match	8.3.667
ㅤㅤㅤNET MAUI Android API	8.3.947
ㅤㅤㅤNET MAUI Android Core Basic	8.3.666
ㅤㅤㅤNET MAUI Android Core Match	8.3.667
ㅤㅤㅤNPM API	8.3.1243
ㅤㅤㅤNPM Core Basic	8.3.681
ㅤㅤㅤNPM Core Match	8.3.682
ㅤㅤㅤFlutter API	8.3.1290
ㅤㅤㅤFlutter Core Basic	8.3.633
ㅤㅤㅤFlutter Core Match	8.3.634

Component	Version	Reference
Face Comparison	`regula-002`	NIST report
Face Age Estimation	`regula-000`	NIST report