Composing a performance

The layers

The performance is built on a layered timeline. Each layer has its own track, and all layers play together — blending into one coherent performance rather than playing one after another. From the bottom up:

• Voice — the dialogue, as one or more voice clips.
• Body gestures — expressive movements like a shrug, a nod, or the sunglasses move.
• Facial expressions — looks like a smile, a raised eyebrow, or unimpressed.
• Head look — where the head is aimed.

Lip sync is solved automatically from the voice and rides along in the same stack.

Choosing and placing clips

Open a track's clip picker and hover over a clip to preview it on the avatar — you see exactly what a motion does before committing. The library is drawn from the campaign footage: slow nods, considering looks, shrugs, sunglasses moves, in short and long variants.

Add a clip to its track and position it in time. Directing is a matter of placement and timing — choosing which moments happen, and when — while the studio handles all the blending between clips. There is no keyframing of the character.

Clips on every track can be dragged to reposition (they need free room to move into) and trimmed by their edges to tighten timing.

Expressions in the final render

Smiles come through the AI render strongly. Subtler expressions — eyebrow raises and the like — can read softer in the finished image than they look in the preview. You can still use them; just preview your renders with this in mind.

Head look

Place a look clip and drag the on-screen target to aim the head — the avatar follows live as you drag. The head turns for the duration of the clip and returns to neutral when the clip ends.

To make the avatar look one way and then directly another way — without returning to neutral in between — place two look clips back-to-back. To have him return to neutral first, leave a gap between them.

The record region

The bar at the top of the timeline sets which part of the timeline becomes the video — drag and trim it like any clip. Clips are short by design: up to about ten seconds. Clip length directly affects render time, so keep clips as long as they need to be and no longer.

Scrubbing and playback

The timeline shows the voice as waveforms with a moving playhead and a time readout. Click or drag anywhere to jump to that moment — the avatar's pose, expression, and framing update instantly. This is where most directing happens, and it costs nothing.

Playback is controlled with play / pause / stop; stop returns the playhead to the start. Scrubbing works while stopped or paused.

Choupette

Choupette — the avatar's cat — can be toggled on and off in the settings, appearing on the shoulder. A miaow can be triggered, her look direction set, and a handful of expressions chosen.

Note

Choupette runs on a newly integrated model that is still being refined — feel free to experiment with her, but expect the occasional rough edge for now.

Screenshot placeholder

Timeline with clips across the layers, and the hover preview (to be added).