Skip to content

1.0.9

Choose a tag to compare

@WengYuehTing WengYuehTing released this 16 Oct 12:05
· 7 commits to main since this release

What's New

  • Migrated real-time audio and video transport from VolcEngine RTC to Agora RTC, with dedicated X2.1.1-preview models for China and US service endpoints.
  • Added independent uplink and downlink quality adaptation based on network conditions, with one-hour quality history and resolution-aware bitrate budgets.
  • Preserved the original aspect ratio of local image and video inputs while selecting the model generation canvas separately and using server-side contain padding.
  • Fixed external video encoding settings so the active RTC connection receives the requested resolution, frame rate, and bitrate.
  • Added uplink and downlink statistics to XLab, plus Debug-only quality tier controls that remember successful selections while keeping automatic adaptation active.
  • Added face restoration for generation requests and enabled it for XLab's virtual try-on mode.
  • Improved front-camera mirror consistency and SDK log readability.

API Changes

  • Breaking: Replaced the previous model constants with RealtimeModel.X2_1_1_PREVIEW_CN (x2.1.1-preview-cn) and RealtimeModel.X2_1_1_PREVIEW_US (x2.1.1-preview-us). Update model selections when upgrading.
  • Added faceRestoration: Boolean = false to both startGeneration overloads. Recompile callers against the new method signatures; Kotlin calls that omit the parameter keep the default behavior.
  • Added setUplinkVideoStatisticsListener and setDownlinkVideoStatisticsListener, with UplinkVideoStatistics, DownlinkVideoStatistics, and their listener interfaces. The reported RTC end-to-end delay is not a measurement of the complete AI generation round trip.
  • Made audio and SDK-hosted preview capabilities optional for RealtimeExternalVideoSource, enabling video-only custom camera capture without implementing audio or preview methods.
  • Local image and video videoFormat values now describe the media input and must preserve its aspect ratio. When omitted, the SDK uses the orientation-corrected, even-aligned media dimensions and model default frame rate; the model canvas is resolved separately.
  • Replaced the runtime RTC dependency with io.agora.rtc:full-rtc-basic:4.6.4. Manual integrations must replace the VolcEngine dependency with Agora; Maven integrations resolve it transitively. XmaxSDK no longer requires the VolcEngine Maven repository or Jetifier.