[navigation]
Key takeaways
- Banuba Face AR SDK tracks 68 facial anchor points and reconstructs a 3D face mesh with up to 3,308 vertices using 37 morphs, so effects stay locked to the face instead of sliding during motion.
- It tolerates up to 70% facial occlusion and holds tracking across a -90° to +90° head-angle range, which is what a live stream actually looks like.
- b.stage reached 1 million monthly active users in two years using Banuba face touch-up and 3D masks in its live streaming and 1:1 chat.
Why does engagement stall in creator apps?
Creator apps compete on how long people stay in the camera, not on how many filters ship. Two things break that. The first is fidelity: a mask that drifts a few pixels during a head turn reads as cheap, and users stop opening the camera tab. The second is catalog depth, because the same six filters get exhausted in a week.
Banuba's face AR SDK addresses the first with a patented 3D Face Kernel that reconstructs geometry directly rather than estimating depth from 2D landmarks, keeping the effect anchored when a creator moves fast or is partly out of frame.
What does the tracking engine actually give you?
The engine tracks 37 facial morphs instead of hundreds of static points, which keeps CPU load down on the mid-tier Android hardware most creator audiences are on. From those morphs, it produces:
- 68 facial anchor points and a 3D face mesh with up to 3,308 vertices, reconstructed directly rather than converted from 2D.
- Tolerance for up to 70% facial occlusion, so an effect survives a hand, a microphone, or hair across the face.
- A -90° to +90° head-angle range, with tracking that stays locked at profile angles and in low light.
- Up to 5 simultaneous faces, which matters for co-streams, duets, and group chats.
Banuba's Face AR SDK filters example
Banuba's published performance table records 30 FPS for single-face tracking on iOS and high-tier Android and 25 FPS on low-tier Android, measured under lab conditions. Face detection distance in the same table is 170 to 180 cm on Android and 230 cm on iOS, which covers handheld and arm's-length framing.
Catalog depth is the second half. The Banuba Asset Store carries over 1,000 ready-made AR filters, and Banuba Studio lets a team build custom effects in-house rather than commissioning each one.
What does the integration look like?
Banuba Face AR SDK supports HTML5, iOS, Android, Windows, macOS, Unity, Flutter, and React Native from a single integration, running on iOS 13+ and Android 8.0+. For creator and gaming teams already in Unity, the Unity integration guide covers the setup, and the quickstart-unity repo is a working project rather than a snippet. There is also a documented sample showing the SDK running alongside Agora for AR filters inside live video calls, a setup live-streaming teams ask about often.
Banuba Face AR SDK works offline just as well as with an internet connection, so tracking and rendering run on the device. That removes the cloud latency budget and the privacy review that usually follows it. If you are weighing this against other options, this hands-on comparison of face filter SDKs walks through the same trade-offs from a developer's side.
Which creator platforms have shipped this?
bemyfriends built b.stage, a fandom SaaS platform whose owners include T1, Sentinels, Hwasa, and Chungha, for live streaming and 1:1 chat. It chose Banuba for natural-looking touch-ups that do not blur skin texture, the size of the effect library, and support responsiveness. In May 2024, b.stage passed 1 million monthly active users two years after its April 2022 launch, with users from 224 countries and over 130 customers globally.
Banuba's AR effects in Bermuda app
Two other deployments in the same shape: Bermuda, a live chat app, has passed 20 million downloads with Banuba effects used over 15 million times a month, and Clash of Streamers reached 4 million installs and 100,000 monthly active users.
Try it on your own build
The fastest way to judge tracking quality is to point it at a moving face on your own target device. Start a free trial of Banuba Face AR SDK, or book a demo to walk through the integration path for your stack.
-
Banuba offers two licensing models. A flat annual license fee gives unlimited usage and a predictable cost that stays fixed even if the app scales to a large user base. An Active User Based model tracks the number of active users over time, which means significantly lower spend early on but a bill that grows with the user base and can exceed the flat fee if the app takes off. The Face AR SDK pricing guide explains how each is calculated.
-
Banuba publishes working sample projects rather than snippets: iOS samples, Android samples, and quickstart-unity for Unity teams. Each runs as-is once you add a client token.
-
Banuba maintains a public AI skills repository that gives coding agents the context they need to wire the SDK into an existing project, which shortens the first-integration pass for teams working with AI assistants.
-
The mobile SDK is around 15 Mb, and the exact figure depends on which feature set you enable. Banuba documents this in its SDK size FAQ.
-
No. Banuba Face AR SDK works offline just as well as it does online, which means face tracking and effect rendering run on the device itself. The Face AR SDK documentation covers the processing pipeline and platform requirements in full.
-
Banuba Face AR SDK tracks up to 5 simultaneous faces with acceptable quality on most mobile devices, with the real ceiling set by device capability and screen proportions. That covers co-streaming and group video chat, the two formats where Banuba's face filters tend to drive the most repeat sessions.
