How it works

From one photo to a live agent on your site.

Four steps, no developer. The first three happen in your dashboard; the last is one line of code on any page.

01 · Photo

Start with a face.

Upload a head-and-shoulders portrait, or make one in the Studio from a description. The background stays by default; remove or replace it if you like. If the photo is of a real person, they sign a short likeness release first.

02 · Avatar

Built in about a minute.

We cut the subject out, build the idle loop (breathing, blinking, small movements) and the speaking model. Preview it saying a line in the voice you chose before you go further.

03 · Agent

Give it a job.

A name, a persona, the files it should know (PDFs, docs, price lists), whether answers stay inside those files, whether it may use the visitor’s camera, a greeting, and limits. Test it live in the builder; the preview runs the real stack.

04 · Embed

One line on your site.

<script
  src="https://cdn.avaclone.ai/v1/widget.js"
  data-agent="pub_…"
  async></script>

A launcher bubble with the avatar’s idle loop; a click opens the conversation. The snippet carries a publishable id only, works only on the domains you allow, and can be revoked.

What a visitor sees.

The panel opens over the page: the avatar, live captions, a microphone, a camera toggle, and a text box for people who would rather type. It is the visitor’s choice to turn the microphone or camera on; both show an indicator while they are, and one tap turns them off.

The AI badge is always on the avatar. Asked “are you real?”, it says no. Every avatar on this page is AI-generated.

Harbor Dental

Book a cleaning in two minutes.

Same-week appointments, insurance checked while you wait.

Cleanings

Whitening

AI avatar · not a real person

Visitor Do you take Delta Dental?

Avatar We do, in network. New patients are seen within the week.

Listening

What the agent can do in a conversation.

Everything below is inside the minute price; nothing is an add-on.

knowledge

Answers from your documents

Upload what the business knows. In “knowledge only” mode the agent answers from those files and nothing else; in “mixed” mode it may add general knowledge. Questions it cannot answer are listed for you in the dashboard.

camera

It sees what a visitor holds up

A part, a label, a room, a building. The agent identifies it, reads any text on it exactly, and looks it up in your catalogue before answering. It never reads people: no emotions, no faces, no identities. That is a product rule, not a setting.

voice

Listens and talks like a call

Speak or type. The avatar listens while you talk, can be interrupted mid-sentence, and keeps the thread. Stock voices on every plan; voice cloning and expressive voice on Pro and above.

Measured, not promised.

0.3 seconds to a live face

A warm avatar is on screen 0.3 s after the visitor connects, measured on our GPUs. Until then the idle loop plays from the CDN: the widget is never blank and never silent.

Three conversations per GPU

Our standard engine runs three live conversations on one card with no late frames. That is what keeps a minute at the price on the pricing page.

Never mid-conversation

Publishing a new version of an agent swaps it for the next session; a visitor already talking finishes on the version they started with.

Fifteen free minutes, no card.

Build one avatar and one agent, put it on a page, and talk to it. Plans from $49 a month when you are ready.

See pricing Start free