Improved handling and reporting of network request issues (e.g., session authentication, model loading).
The SDK now performs a consistency check in Electron apps using reference data after initialization. If inconsistencies are detected, it will automatically fall back to the WASM backend.
Native inference modules are available for Electron, offering better performance and lower resource usage compared to the WASM backend. See documentation for more details.
Improved resource caching logic.
Enhanced precision in getMetrics, especially during the first second of execution
Automatically falls back to the CPU WASM provider if an error occurs in the WebGPU provider during runtime.
Added more details for onError callback
Introduced the ability to dynamically load effects into the pipeline.
A new method has been added to retrieve various metrics, including: Output FPS, Segmentation inference time, Full frame draw time (including segmentation inference)
Removed unnecessary allocations and copies for improved integration with Electron native modules.
Optimized the camera stream switch for faster performance.
Fixed the background positioning issue when rotating the device between portrait and landscape modes.
Resolved the problem with incorrect output stream resolution when SDK is not enabled (impacting Safari on mobile devices).
Reduced CPU Overhead on WebGPU: Significantly optimized CPU usage during segmentation tasks when running on WebGPU, ensuring smoother performance and better resource efficiency.
isSupported API Endpoint: Introduced a new isSupported API to help developers easily check for compatibility with the SDK before initialization.
Multi-Instance GPU Segmentation Support: Added support for running multiple SDK instances with GPU-based segmentation on the same page, enabling more flexible use cases.
NPM Package Enhancements: Improved the NPM package structure by providing separate files for ECMAScript Modules (ESM) and CommonJS, making it easier to integrate into different project setups.
GPU Device Lost Handling: Implemented robust handling for GPU device loss scenarios, particularly addressing issues on older Intel Integrated Graphics drivers.
Dynamic Eval and CSP Compatibility: Fixed issues related to dynamic eval and script loading, improving compatibility with stricter Content Security Policies (CSP).
Worker-Timer Errors in Firefox: Resolved Firefox-specific worker-timer errors by moving them to the ErrorBus system instead of throwing exceptions, ensuring smoother error handling.
Background Blinking on Intel Integrated Graphics: Fixed an issue where background blinking occurred during parallel segmentation tasks across different tabs on Intel Integrated Graphics.
Person Extraction with Transparent Background: Added the ability to extract a person with a transparent background (available in canvas and MediaStream).
Mirroring Effect: Introduced a mirroring effect that mirrors the original stream and applies all subsequent effects to the mirrored stream, eliminating previous issues with orientation and alignment.
Beautification Effects: Optimized beautification effects to reduce GPU usage by 50%, enhancing performance.
Granular Effect Control: Added more detailed control over which effects are included in the main processing pipeline, providing better customization and stability.
Third-Party Library Optimization: Removed third-party libraries that used eval, leading to a more secure and optimized structure.
Memory Leak Resolution: Fixed a memory leak issue in frame bypass mode for Safari and Firefox, improving overall stability and performance on these browsers.
New Feature: Low Light Correction - This enhancement, also known as low light adjustments in Zoom, significantly improves video quality in suboptimal lighting conditions or when using lower-quality cameras.
Enhancements in Stability and Speed - Various optimizations have been implemented to enhance the overall stability and speed of the software, ensuring a smoother and more reliable user experience.
Take a look at our demo applications for Windows, macOS, iOS, Android and the Web. Get an understanding of how our AI effects work in real environments.