Google Meet, Epic Games Use New LiteRT Framework for On-Device AI
Google has detailed LiteRT, a cross-platform framework for accelerating on-device AI workloads by utilizing Neural Processing Units (NPUs). The framework is designed to help developers deploy features like real-time video effects and animation without compromising performance or battery life. Early adopters include Google Meet, which uses it for ultra-HD background segmentation, and Epic Games, for real-time facial animation streamed into Unreal Engine.
Key Takeaways
- Google Meet is using LiteRT to power an ultra-HD background segmentation model that is 25x larger than before, without increasing power consumption.
- Epic Games’ Live Link Face app for Android uses the framework to achieve up to 30 FPS for real-time MetaHuman facial animation.
- LiteRT provides a unified API, abstracting vendor-specific SDKs to target NPUs on mobile (Qualcomm, MediaTek), desktop (Intel), and IoT platforms.
- In performance tests, speech recognition firm Argmax saw a >2x speedup by shifting workloads from GPU to NPU using LiteRT.
Why It Matters
The framework directly addresses the performance and battery life tradeoffs of running sophisticated AI models on client devices for real-time video. By abstracting the hardware layer, LiteRT could standardize how developers deploy features like advanced background replacement and real-time animation across a fragmented Android ecosystem. The key signal to watch will be adoption by other major video-centric applications and SDKs, which would indicate if LiteRT is becoming a de facto standard for accessing on-device NPUs.
Read full article at developers.googleblog.com
Enjoy our coverage?
Add StreamingMeme as a preferred source on Google to see more of our streaming news at the top of your Search results.
Add as preferred source