Avatars and voices

Choosing 2D or 3D

Pick 2D when the avatar represents a real, recognisable person, or when photorealism is the point — a government spokesperson, a brand ambassador, a named expert. It is built from studio footage of a real human, so it looks like that human because it is that human.

Pick 3D when you want a character rather than a person: a mascot, a stylised guide, a figure that has to move around, change outfits, gesture, or exist in a scene. Unreal Engine 5 gives you animation freedom that footage cannot.

3D also has a decisive practical advantage: it can render on the end-user's device, including a smartphone. The GPU cost per concurrent user drops accordingly, which matters when the avatar lives in a consumer app rather than on hardware you own.

The avatar catalog

Every deployment gets a catalog of avatars addressed by a stable id — for example f7f6fe47f93446c1. That id is what you pass to the session API or the embed URL. Catalog contents are per-customer: your custom avatars are visible to your deployment only.

Building a custom avatar

2D photoreal needs a short, controlled recording session:

• A few minutes of frontal footage of the person, neutral expression, even lighting, static camera, mouth clearly visible.
• Written consent from the person being captured — we require it before processing, and it belongs in your records too.
• Turnaround is typically a few days from receiving usable footage.

3D starts from either a supplied character mesh or a photo-based reconstruction, then goes through rigging and look development in Unreal Engine. Turnaround depends on how much art direction you want.

Both paths end the same way: a new id in your catalog.

Voices

The voice is configured per avatar, independently of the face. You can:

• Pick a stock voice in the target language.
• Use a commercial voice provider you already have a contract with.
• Clone a voice from a recording, given documented consent from the speaker.

For Arabic deployments we generally recommend auditioning at least two voices with your own script before committing — dialect and prosody preferences vary more than the sample clips suggest.

Branding around the avatar

The surface around the avatar — logos, colours, background, subtitles, idle behaviour — is configurable per deployment. Kiosk builds also control screen rotation, attract-loop behaviour, and the on-screen language switcher.