ACE Unreal Plugin Changelog

2.4.0

• New features

  • Six new plugins which execute Audio2Face-3D locally using an NVIDIA GPU: NvAudio2FaceClaire, NvAudio2FaceJames, NvAudio2FaceMark, NvAudio2FaceClaireRegressive, NvAudio2FaceJamesRegressive, NvAudio2FaceMarkRegressive

  • Audio plays back at original quality from Audio2Face-3D

  • Audio2Face-3D real-time input mode, which will send audio into inference in real-time instead of bursting it as quickly as possible

  • Application-provided runtime hint to pre-allocate Audio2Face-3D resources

  • Application-provided runtime hint to free Audio2Face-3D resources early

  • Support streaming Sound Wave assets, for example “On Demand” loading behavior, as input to Audio2Face-3D

  • Runtime Animation Stream (NVIDIA Animation Graph) server URL override

  • Animation Stream connection timeout, project setting and command line override

  • Support receiving audio streams of arbitrary sample rate and number of audio channels from Animation Stream

• Fixes

  • Fix crash on exit if audio samples are still being sent to Audio2Face-3D server

  • Fix memory leak when receiving animation samples

  • Fix possible runtime errors with the Audio Send Completed action for Animate Character async blueprint nodes.

  • Allow blend shape weights outside the range [0.0, 1.0] for the rare cases where that’s needed

  • Fix possible missing animation or deadlock on application exit when sending audio into Audio2Face-3D from multiple sessions at once

  • Fix main thread getting blocked by background Audio2Face-3D work

  • Fix issue where the application calling AnimateFromAudioSamples multiple times for the same session was unsupported in the RemoteA2F provider

  • Correctly return failure instead of success when Animation Stream connection fails

  • Fix instability due to GPT local execution feature being created on unsupported hardware

  • Fix issues with long path support for GPT local execution on Windows

• Other changes

  • Deprecate old non-async Animate Character blueprint nodes

  • Improve MetaHuman pose asset (mh_arkit_mapping_pose_A2F)

  • Protect against applications deleting an Audio2Face-3D session from one thread while sending audio into the same Audio2Face-3D session from another thread, which could otherwise result in crashes or other issues

  • Protect against applications sending audio into an Audio2Face-3D session from multiple threads simultaneously, which could otherwise result in mixed up speech samples and cacaphony

  • Protect against applications attempting to end an Audio2Face-3D session that doesn’t exist

  • Add more logging on Animation Stream connection errors

  • Prevent running multiple RemoteA2F provider sessions simultaneously, which was not working properly

  • Prevent running multiple instances of GPT local execution simultaneously, which was not working properly

  • Add support for Unreal Engine 5.5, drop official support for Unreal Engine 5.3

2.3.0

• New features

  • GPT local execution

  • Optional custom blend shape weight multipliers and offsets for the “Apply ACE Animation” animgraph node

  • Blueprint function to query blend shape names: “Get Default Blendshape Map”

  • Default NVIDIA Audio2Face-3D provider “RemoteA2F” supports Linux and Win64. The old Win64-only “LegacyA2F” provider is still available.

  • Add latent blueprint nodes to start animations from Audio2Face-3D without blocking

  • Add C++ API to send raw audio data into ACE Unreal plugin

• Fixes

  • Fix connections to Animation Data servers using https crashing on Linux

  • Fix samples possibly not being sent for some Audio2Face-3D sessions (issue introduced in 2.2.0)

• Other changes

  • Update MetaHuman pose asset (mh_arkit_mapping_pose_A2F) to support new Audio2Face-3D v2.3 models

2.2.0

• New features

  • “Animation Stream” feature allows facial animations to be driven by the ACE Animation Data v1.0 service (also known as Animation Graph)

  • Linux support for applications and Unreal Editor (Animation Graph only, no Linux Audio2Face-3D support currently)

  • Animation start and end events

• Other

  • Plugin now supports selecting the Audio2Face-3D provider at runtime, although there is currently only one provider, which is identical to the previous release

  • More logging (Verbose or VeryVerbose) for diagnosing animation synchronization issues

2.1.0

• New features

  • Connect to NVCF-hosted Audio2Face-3D service via API key

• Other

  • Improve logging for some BP functions when inputs aren’t supplied

2.0.0

• New features

  • Initiate an Audio2Face-3D session by sending an audio clip to the service directly from an Unreal application

  • Spatialized audio

  • Support source audio clips of any sample rate

  • Support Audio2Face-3D v1.0 service

  • Application control of emotion information used to generate facial expressions

  • Override Audio2Face-3D server face parameters

• Fixes

  • Fix audio and animation synchronization issues. Manual delay/offset is no longer needed

  • Fix audio not supported in certain cases, such as when using Pixel Streaming

  • Fix multiple character support

• Other changes

  • Deprecate LiveLink support

  • Add support for UE 5.4, drop support for UE 5.2

1.0.1

• Rename plugin from “OmniverseLiveLink” to “ACE”

• Add UE 5.3 support

• Write plugin version to log file

1.0

• Initial “OmniverseLiveLink” plugin release