Avatar

An avatar is a live face for your agent. On an in-app or website call, the face appears where the agent's picture usually is and speaks with the agent's own voice, its lips in sync with every word. Everything else about the call stays as it is: the same agent, the same captions, the same actions and components on the customer's screen.

Phone calls never have a face.

Turn it on

In the dashboard, open the agent, then Embed & Widgets → Appearance → Avatar.

1. Turn on Show a live face on calls.
2. Add a Portrait: paste a link to the photo (it must start with https://), or choose Upload to add one from your computer. A preview under the field shows the portrait as it appears on a call screen: a 4:5 card with your agent's name, AI assistant and a red LIVE across the top.
3. Choose Adjust framing to centre the head and shoulders. See Framing the portrait.
4. Quality: Standard is smooth on every connection and right for most agents. High is the most lifelike face, with sharper detail and fuller movement, for agents where the face matters most and users on good connections.
5. Optionally, add a Background: a link to an image, or a short video, shown behind the face. Leave it empty to keep the portrait's own background.
6. Leave Wait for the face before the greeting on, or turn it off to have the agent greet at once. See On a call.
7. Choose Save.

The change applies from the next call. On an agent that is live and needs approval for changes, the Avatar setting waits for an Admin's approval like any other change to the agent.

To turn it off, turn off Show a live face on calls. That is saved at once; the next call is voice only.

Choosing a portrait

The face is made from one photo, so the photo decides how it looks.

• One person, facing the camera, head and shoulders in the frame.
• Even light on the face, with no strong shadows.
• Nothing covering the face — no sunglasses, masks or hands near the mouth.
• A PNG with the background removed if you want your own background to show behind the face. A photo with its background keeps it.
• PNG, JPG or WEBP, up to 2 MB. A larger photo is resized before it is uploaded.

Framing the portrait

The face looks best when it is made from a square photo with the head and shoulders in the middle. A face that fills too little of the picture has less room to move, so lips, blinks and teeth look stiffer.

Graine frames every portrait for you on the call: it finds the face and brings it to the size that looks best, from the full size of your photo. A sharp photo, at least 1024 pixels on its short side, keeps the face sharp after that. To choose the framing yourself, use Adjust framing and fill more of the square with the head and shoulders. A picture with no face in it is left as it is.

1. Next to the preview, choose Adjust framing. The photo opens in a square frame with a faint guide: an oval for the face and a line for the shoulders.
2. Drag the photo to place the face in the oval and the shoulders on the line. Drag with the mouse, or with a finger on a touch screen; the arrow keys move it too. The photo always fills the frame: it stops at its edges.
3. Use Zoom, from 1× to 3×, to bring the face closer.
4. Choose Use this framing. The framed square is uploaded as the portrait, 1024 × 1024 pixels, and the preview shows it. It is a PNG when the photo has transparent areas, so your background still shows behind the face, and a JPG otherwise.
5. Choose Save.

Cancel closes the frame and leaves the portrait as it was.

A photo you linked from another website can be framed only if that website allows other pages to use its images. If it does not, the panel says To adjust the framing, upload the photo instead of linking it. Choose Upload to add the photo from your computer, then frame it.

On a call

• The agent's first words wait for its face. With Wait for the face before the greeting on (the default), the greeting waits until the face is on screen, about two seconds at most. If the face is not there by then, or cannot be shown, the agent greets with its voice alone. A call without the face greets at once.
• With the switch off, the agent greets at once. The face appears a moment after the call connects; until then the agent's usual picture shows and the agent can already be heard.
• The voice moves to the face at a pause, never in the middle of a sentence, so nothing is heard twice.
• If the connection to the face drops, the agent's voice carries on at once and a short line says Video paused while the connection recovers. You can keep talking. The face comes back by itself when it can.
• If the face cannot be shown — the network does not allow it, or every face is in use at a busy moment — the call goes on with the voice alone, and the customer sees Video isn't available right now. Your call continues with voice.
• Captions, actions and components are unchanged. Interrupting the agent works exactly as it does without a face.

The face plays over the same kind of secure, encrypted connection as the call. On networks that block everything but web traffic, it connects over port 443.

Where the face appears

What it costs

Minutes with the face on screen are charged on top of the call, per minute, to the second. A call without the face — because it is turned off, or because the face could not be shown — is charged as a normal call.

Your workspace's price is shown under the switch. In Billing → Usage, these charges appear as Avatar minutes.

Privacy

• The portrait you upload is stored by Graine and served from a link, like your widget's logo: anyone who has the link can open the image. Do not upload a photo you need to keep private.
• The face is drawn from the agent's own speech. The customer's microphone and camera are never used to draw it, and their camera is never opened.
• The face is drawn live for the call. The call's recording and transcript are the same as without a face.