Use cases
Where an on-device face actually changes the product.
Not every avatar belongs on the device. These are the shapes where it does — long sessions, high engagement, or a network you can’t rely on.
Companions and characters
The users you want most are the ones who talk for an hour. On a per-minute plan those users are a liability; here a heavy user and a light one cost the same to render. This is the single clearest fit.
Language tutoring
Daily practice means daily minutes, and a tutor that pauses to buffer breaks the illusion instantly. No round trip means no buffer, and a commute with patchy signal stops being a problem.
In-app support and onboarding
Support volume spikes exactly when you least want a per-minute bill. Rendering cost stays flat through the spike because it never touches your infrastructure.
Games and NPCs
A talking character per player does not scale on rented GPUs. Every player brings their own renderer, so a hundred thousand simultaneous conversations cost the same as one.
Health, coaching and journalling
Sensitive conversations where no audio, video or transcript leaving the device is a compliance argument as much as a technical one.
Kiosks and field devices
A face that keeps working when the venue wifi does not. Rendering carries on through a dropped link; only the conversation model needs reconnecting.
Where we’d tell you to use someone else
Short one-off interactions on the web, where a three-minute session never accumulates enough minutes to matter and a browser is the whole product. Anything that needs Android today. And any full-body character in an arbitrary scene — we render a face, not a world.