When to generate previews, and why they need updating

Making an avatar speak on screen is a separate process from preparing its audio.

A still presenter during editing is not necessarily an error. First listen to the voice, pronunciation and speaking speed. Once these are reasonably settled, generate pending previews to create the moving face and mouth. This reduces unnecessary video regeneration. Preview generation is also separate from exporting the finished MP4.

Tracks hold instructions for different roles along the timeline.

Speech specifies what is said and when; base expression sets the sustained expression; emotes add reactions at chosen moments. Head motion and gaze are also linked to the target avatar. Preview generation turns these instructions into video for the avatar-clip interval. Simply adding an empty track does not necessarily require regeneration.

Check For a conversation, check which presenter each track belongs to so that speech and reactions affect the intended person.

A generated video reflects the voice and facial settings used at the time.

When speech, voice, timing, base expression, gaze, emotes or head motion changes, the old video no longer matches the instructions. The affected avatar clips need updated previews. Even identical text can require new mouth movement if the audio is regenerated. Background replacement, moving a presenter within the frame and changing caption colors usually affect composition instead.

Check Moving a presenter within the picture is different from changing a clip’s start or end time. After changing the interval, check its generation status.

The more speech a long avatar clip contains, the more a small revision can affect.

If one avatar clip contains three speech clips, editing the middle line updates that avatar clip’s video. Split speech at meaningful boundaries, then make avatar clips cover each speech unit and its surrounding pauses. This helps limit revisions. Some edits, such as changing a track-wide baseline expression, can still affect multiple clips.

Check Shorter intervals help narrow the work to revise. Waiting time also depends on the type of change and service load, not only clip duration.

A consistent review order makes editing easier.

Listen first and settle the script, pronunciation, voice and pauses. Then adjust base expression, emotes and head motion. Select Generate pending previews and check the displayed target count and progress. When editing a translated version, preparation may include translation and audio. Wait for the operation to finish rather than repeatedly submitting it.

↘Red arrows and white labels are guide annotations ＋ Enlarge with callouts

1. 1
   Action Click Generate ungenerated clips.
2. 2
   Check After generation, the avatar clip will say Preview generated.

To review a single clip

Select the avatar clip and use Generate this clip. This is unnecessary when the clip is already generated with the latest settings.

Intervals without speech

Silent intervals, such as a listening presenter, can also be generated. No speech does not always mean that no avatar preview is needed.

Check the actual voice and movement as well as the generated status.

Move the playhead to just before the edited interval and play it. Check lip sync, expression and emote timing, and transitions at clip boundaries. If the presenter remains still, check generation completion, clip status, playhead position and whether the track is enabled. Once the preview looks right, export an MP4 and review the exported video too.

↘Red arrows and white labels are guide annotations ＋ Enlarge with callouts

1. 1
   Action Click near the beginning of the ruler.
2. 2
   Action Click the triangular Play button.
3. 3
   Check Check Saved.