Audio

Audio Workflow

Audio drives facial animation from a voice track alone — no footage, calibration, or identity. It is the simplest job type to configure, for lip-sync and expression from dialogue when no capture footage exists.

Four Audio jobs in the queue, typed Face / Auto Detect, running through Tracking to Done with waveform thumbnails
An audio batch — one job per voice track, no footage or identity needed

Pipeline stages

Validate Input
->
Audio SolveVoice -> controls
->
Export Outputs
->
Save Assets

Required input

One voice track per job, from any of three sources in the Add Jobs... dialog:

SourceTypeNotes
Project SoundWave USoundWave A sound wave asset already in the content tree
WAV folder Files on disk The WAV stays where it is; each job imports it beside the Performance at solve time
Wwise container Resolved over WAAPI Solves from Wwise's own converted media; conversions must be PCM. See Wwise Audio
Audio jobs do not require Capture Data, a MetaHuman Identity, or a Camera Calibration — a voice track is the whole input.
Audio solves need UE 5.8. The solve runs on 5.8-only Performance APIs, so on UE 5.7 an audio job fails immediately with a version message; the rest of the plugin works normally.

Audio options

Solver Mode Realtime — faster, lower quality, good for previews and iteration. Offline — slower, higher quality; use for final deliverables.
Mood Preset Emotional baseline the solve colors the animation with: Auto Detect (the default, inferred from the audio), Neutral, Happy, Sad, Disgust, Anger, Surprise, Fear, Confident, Excited, Bored, Playful or Confused. It applies to the offline solve; the realtime solve has its own Realtime Mood pair, defaulting to Neutral.
Mood Intensity Blend weight for the Mood Preset — lower values keep the animation closer to neutral while retaining a hint of the chosen mood.
Output Controls Which curves the solve outputs: Full Face (all facial curves) or Mouth Only (lip sync curves only).
Generate Blinks Epic's procedural blinks. Off and ignored while Natural Blinks is on, which it is for new Audio jobs.
Downmix Channels / Audio Channel Downmixes multi-channel audio to mono by default. Disable to solve from a single channel, then set Audio Channel (0-indexed — 1 for dialogue on the right channel).
Lookahead Window How far ahead (ms) the solver reads the signal for anticipatory cues, e.g. jaw pre-opening before a vowel. Default 80 ms — raise for more anticipation, lower for tighter sync.

Eyes

The Eye Director adds procedural eyes to offline, Full Face audio solves. It is applied at export (AnimSequence, Level Sequence, FBX); the solved Performance on disk stays the raw solve.

Natural Blinks Blinks at a natural rate, clustered on speech pauses, in place of Epic's. On for new Audio jobs.
Gaze Director Holds one gaze direction for the take, with fixation flicks and aversions at phrase starts. Off by default.
Gaze Direction Where the eyes rest, -1..1 per axis (default 0, 0); X > 0 is the subject's left. Drag the picker or type it.
Gaze Aversion How far fixations and aversions wander, 0..1 (default 0.5). 0 is a steady stare.
Gaze LookAt Target Aims the eyes at a GazeTarget in the exported Level Sequence, blended by the LookAt Weight strip. Off by default. See Level Sequence Preview.

In the Performance editor the rows sit directly below Generate Blinks in the Details panel (Processing Parameters): a live blinks / aversions status line, Natural Blinks, Gaze Director, Gaze Direction with its picker, Gaze Aversion, Gaze LookAt Target, and the LookAt Weight strip. Ticking Gaze LookAt Target regenerates the weight keys from the speech; undo restores the old keys. The viewport LookAt marker hides while the weight is 0. The job settings panel in the MHA Manager tab has the same rows plus a Gaze Direction picker.

The settings travel with the Performance asset, through rename and save, but not through Duplicate or Save As - a copy starts untuned. Realtime and Mouth Only solves carry no eye curves, so the rows hide for them.

Outputs

OutputWhenNotes
Performance + AnimSequence Always Solved performance plus baked face animation, driving full-face or mouth-only controls per Output Controls
FBX exports Optional Enabled in Global Settings; face jobs get the Control Rig _CR form