The Panel

Character: faces, recognition & publishing

How the character is drawn, whether it recognises the person in front of it, and how you put it in front of people.

3 · LiveFace 2D — a talking head from one photo

LiveFace 2D tab with the render target switch, engine selection cards and the avatar upload area
LiveFace turns a photograph into a face that speaks your character's words.

LiveFace renders the character as a real photograph whose mouth moves with the speech, instead of a 3D model. It is the fastest way to a convincing face: one good headshot and you have a talking presenter.

EngineTrade-off
LightweightFast, light on the GPU, good quality. The right default.
MuseTalkNoticeably more photoreal, slower, heavier. Worth it for a presenter that will be on screen a lot.
  1. Choose the engine first. The avatar is prepared specifically for the engine you pick.
  2. Upload a photo or a short clip — a well-lit head-and-shoulders shot facing the camera works best. Generation takes roughly half a minute; the first one takes longer.
  3. Switch the render target on.
  4. Press Live preview to check the mouth actually moves.
  5. Copy the embed code and paste it into your site, replacing the placeholder with an API key.
Changing the engine after generating an avatar does not regenerate it — re-upload the photo, or the face will stay frozen. And if no clear face is found in the image, the previous avatar is kept and nothing changes.
LiveFace does not appear in the Playground, which only draws 3D models. Use the Live preview button on this tab to check it.

4 · Pixel Streaming — a cloud MetaHuman

Pixel Streaming tab showing the gallery of MetaHuman avatars available to assign to the character
Pick a MetaHuman and it is streamed as live video from a GPU server — no Unreal install for your visitors.

This gives the character the body and face of a full Unreal MetaHuman, rendered on a GPU server and streamed to the browser as video. Your visitors need nothing but a web page. You are only choosing the look — the voice, the brain, the knowledge and the rules all stay exactly as you configured them on the other tabs.

Click an avatar to assign it, and Disable to go back to the browser-rendered look. The first visitor after a quiet period may wait a little while a GPU server wakes up.

LiveFace and Pixel Streaming are alternatives, not layers — switching one on switches the other off.

13 · FaceID — recognising the person

FaceID tab with the master switch, server settings, matching thresholds, behaviour switches and the greeting script fields
FaceID. Every greeting is written by you — the character speaks these lines word for word.

With a camera attached, the character can tell whether the person in front of it is someone it has met before, and greet them accordingly. Where identity really matters — banking, for example — it can also check a face against the chip in a national ID card, read by a phone.

Behaviour switches

SwitchSet it for
Greet by nameLeave on for anywhere hospitable.
Register by name (ID not required)On for a café or a shop — an unrecognised visitor is simply asked their name and remembered. Off where identity must be proven.
Face match requiredOn for banking: if the ID card checks out but the face does not match, the person is refused.
Send kiosk camera to the cloudOnly when the recognition service has no camera of its own.

The greeting scripts

Seven fields, each spoken exactly as written, with {name} replaced by the visitor's name. Leave one blank and that situation stays silent.

SituationExample line
Recognised face"Welcome back, {name}. Good to see you again."
Unrecognised face"Hello there. I don't think we've met — what's your name?"
ID needed"Could you hold your ID card to the reader for me?"
Verified returning visitor"You're already verified, {name}. How can I help?"
Identity verified"Thank you, {name}. You're verified."
Face didn't match"I'm sorry, the photo on the card doesn't match. Please see a colleague at the desk."
No live face"I can't see you clearly — could you step in front of the camera?"
These fields save when you click away from them. Type a greeting and press Ctrl+S without leaving the box and it is not stored. Click elsewhere on the page first.
Leave the FaceID server field blank only for demos — blank means the shared platform face database. For a bank, a clinic or anywhere with real customers, point it at your own server so faces never leave your infrastructure.

12 · Telefon — give it a phone number

Character telephony tab with inbound settings, phone number, greeting text and the outbound dial box
Inbound settings on the left, a dialer for testing on the right.

Attach a real phone number and the character answers calls in its own voice, with all of its knowledge, memory and tools intact. Full setup is on the Phone calls page; this tab is where each character's own number and greeting live.

Press Save before dialling. The outbound dialer reads the number from the saved record — an unsaved number will not be used as your caller ID.

17 · Publish — putting it in front of people

Publish tab with visibility selector, character ID, Unreal plugin settings, WebSocket snippet and HTML embed code
Publish. Every block has a copy button.
VisibilityWho can talk to it
PrivateOnly you, and only with an API key. The right setting while you are still building.
UnlistedAnyone with the link. The usual choice for a website embed.
PublicEveryone.
Press Publish first, then set the visibility. Publishing sets the character to Unlisted, so if you choose Public and then press Publish, it quietly drops back to link-only. Do it the other way round and check the badge afterwards.

The tab then hands you the three snippets you need:

A publicly embedded character bills your plan for every visitor conversation. That is normally what you want, but keep an eye on Plan & Quota for the first few days after putting one on a busy page.

Where to go next

Set it up to take calls

Phone calls →

Put it on a stream

Going live →

Embed on your site

Web & embed →