Video Avatars for AI Voice Agents

Want your AI voice agent to show up with a face? VoiceInfra Video Avatars add a real-time video avatar to AI voice agents on web calls on the website widget, powered by Anam and Keyframe Labs. Pick an avatar, preview it with your voice, tune its expressions, and publish. Phone calls stay audio-only.

Key takeaway: Video Avatars are a setting on the assistant or workflow you already have. The same prompt, tools, knowledge, and phone routing keep working. The avatar only appears on surfaces that have a screen.

Why Video Avatars, and Why Not on Phone Calls

The Challenge: A Voice-Only Web Experience

A website visitor is already looking at a screen. They are reading a product page, comparing prices, or filling out a booking form. When the voice agent answers without a face, the conversation can feel anonymous, a helpful voice attached to nothing.

The obvious fix is video everywhere. That is usually the wrong fix:

• Phone calls do not need video. The caller is holding a phone to their ear, often while walking or driving. Video adds cost, bandwidth, and a camera nobody asked to use.

• Video on every surface is expensive. Rendering a face for calls that cannot show it burns compute for no customer benefit.

• A face that does not match the brand is worse than no face. Avatars need to be chosen and previewed, not switched on blindly.

• The agent should not fork. Rebuilding the assistant for video creates two products to maintain.

The VoiceInfra Approach: Video Where There Is a Screen

VoiceInfra Video Avatars are web-only by design:

• Portal test calls show the avatar while you build and test the agent.

• Website voice widgets show the avatar during a live call.

• Phone calls stay audio-only. No camera, no video track, no pretending the caller can see something.

The same assistant runs underneath every surface. Turning the avatar on does not change the prompt, tools, knowledge, phone numbers, or call routing. It changes how the agent appears on the surface that can show it.

What You Get with Video Avatars

A Face That Speaks with Your Agent's Voice

Video Avatars are not a separate assistant or a separate product. They are a visual layer on the agent you already configured:

• Avatar catalog: browse avatars from Anam and Keyframe Labs, the providers configured for your deployment

• Style filters: realistic, illustrated, or 3D animated

• Search and facets: find an avatar by name, model, or gender

• Live preview: see the selected avatar before applying it

• Facial expressions: turn expressions on and choose what the avatar can show

• Word-timed movement: the face moves on the word with supported voices

• Full-screen widget stage: the avatar fills the call stage, with conversation and controls on top

• Minimize to a live bubble: keep the call going while the avatar becomes a thumbnail

Built for the Agent You Already Have

• One setting, two surfaces: turn the avatar on for an assistant or a workflow

• No prompt changes: the system prompt and tools stay as they are

• No knowledge migration: the same documents and catalogs remain attached

• No phone changes: numbers, routing, and transfers are untouched

How Video Avatars Work

The Avatar Lifecycle

Open Model & Voice
Turn on Video avatar
Pick an avatar
Preview with your voice
Tune expressions
Publish
Test call
Website widget

Step-by-Step Setup

Step 1: Open Model & Voice

• Open the assistant or workflow in the VoiceInfra dashboard

• Go to Model & Voice

• Scroll to the Video avatar section

Step 2: Turn On the Avatar

• Use the Video avatar switch

• If you have never chosen an avatar, the picker opens automatically

• If you have, the last avatar comes back

Step 3: Pick an Avatar

• Browse the provider catalog

• Search by name

• Filter by style, model, or gender when the provider offers more than one

• Select an avatar to preview it

Step 4: Preview with Your Voice

• The picker shows the selected avatar in a live preview

• The assistant page previews the same avatar beside the voice and expression settings

• Confirm the face matches the voice before you publish

Step 5: Tune Expressions

• Turn expressions on if the avatar supports them

• Choose from the expressions this avatar can show

• Check the voice note: word-timed voices move the face on the word; other voices update it once per reply

Step 6: Publish and Test

• Save or publish the assistant or workflow

• Start a portal test call to see the avatar in the call

• Open the website widget to see the full-screen call stage

What Happens Under the Hood

Every avatar is a video track in the same real-time call as the voice. The audio still comes from the voice pipeline; the avatar is rendered alongside it. Nothing about the agent's prompt, tools, knowledge, or phone routing changes when the avatar is on.

That separation is why one assistant can be audio-only on a phone number and video on a widget. The avatar is a rendering choice, not a second agent.

Avatar Styles and the Picker

Three Style Families

The available styles depend on the avatar providers configured for your deployment. The picker only shows filters that actually distinguish the catalog in front of you.

Search, Filters, and Preview

• Search by name: type to narrow the catalog

• Filter by style: realistic, illustrated, or 3D animated

• Filter by model: when the provider offers multiple avatar models

• Filter by gender: when the catalog includes gender metadata

• Chips on each avatar: provider, style, gender, and model at a glance

• Live preview: the selected avatar stays visible while you scroll

• Remembered choice: switching the avatar off and on brings the last one back

Facial Expressions and Voice Timing

Provider-Driven Expression Settings

Expression controls come from the avatar provider, so they match what the avatar can actually do:

• Turn expressions on or off

• Choose from the expressions this avatar supports

• Options are limited to the avatar's capabilities, not a generic list

Word-Timed vs. Once-Per-Reply

The face can move in two ways:

• Word-timed: when the selected voice reports when each word is spoken, the face moves on the word. ElevenLabs, Cartesia, and Deepgram voices support this.

• Once per reply: other voices update the face once per response.

Word timing is the difference between a face that feels present and a face that feels like a slideshow. If expressions matter to the use case, choose a supported voice.

Realtime Models and Expressions

Expressions are not available with realtime models. Realtime models speak directly, so there is no separate voice step to time the face to. To use expressions, use a separate speech-to-text, LLM, and voice pipeline.

What a Video Call Looks Like in the Widget

The Full-Screen Stage

When a visitor starts a voice call with an avatar-enabled agent, the widget opens a full-screen video stage:

• The avatar fills the stage

• The conversation appears over the video

• Mute, end, and minimize controls sit on the stage

• The visitor can type a message instead of speaking

• The call timer and connection state are visible

Small Details That Matter

• The call button becomes a camera when the agent has an avatar

• The still image holds until the first video frame, so the stage is never blank while the call connects

• Typed lines still appear when the transcript is off, so the visitor's text is never lost

• Cards sit above the text box, so visual answers stay inside the call

• Minimize turns the avatar into a live thumbnail, and the call keeps running

• The avatar face appears on voice widgets, not chat-only or call-request widgets

Beta Limits and Compatibility

What Beta Means

Video Avatars are in beta. That means:

• The avatar catalog and expression controls depend on the Anam or Keyframe Labs provider configured for your deployment

• Phone calls remain audio-only

• Realtime models do not support expressions

• The feature is labeled Beta in the product

Compatibility Matrix

Use Cases for Video Avatars

Website Customer Support

• Put a face on the support widget

• Let visitors see who is helping them

• Keep the transcript and typed input available

• Escalate to a human when needed

Sales and Booking

• Use a realistic avatar as a visible sales presence

• Let the avatar walk a visitor through options

• Book appointments without leaving the page

• Collect form data before the call begins

Product Guidance and Demos

• Show a 3D or illustrated avatar for product tours

• Pair the face with visual cards

• Keep the conversation on the page

• Reduce the need for a separate demo call

After-Hours Web Support

• Cover the website when the team is offline

• Give the visitor a visible, interactive answer

• Capture the details needed for follow-up

• Hand off to a human the next business day

Frequently Asked Questions

What is a Video Avatar?

A real-time video face for an AI voice agent. It renders alongside the voice on web calls, so the visitor sees the agent while it speaks.

Does it work on phone calls?

No. Phone calls stay audio-only. The avatar appears in portal test calls and website voice widgets.

Will the avatar change my agent's behavior?

No. The prompt, tools, knowledge, phone numbers, and routing stay the same. The avatar is a visual layer on the existing assistant or workflow.

Can I choose the avatar?

Yes. Browse the provider catalog, search by name, and filter by style, model, or gender when the provider supports it. Preview the avatar before applying it.

Which voices support word-timed movement?

ElevenLabs, Cartesia, and Deepgram. Other voices update the face once per reply.

Why don't realtime models support expressions?

Because there is no separate voice step to time the face to. Realtime models speak directly. Use a speech-to-text, LLM, and voice pipeline if you want expressions.

Does the widget support avatars?

Yes, on voice calls. The widget opens a full-screen video stage, and the visitor can minimize it to a live thumbnail. Chat-only and call-request widgets do not show the avatar.

What happens while the video is connecting?

The avatar's still image holds until the first video frame. The stage is never blank.

Can I turn the avatar off?

Yes. Use the Video avatar switch. The last avatar is remembered, so turning it back on restores your choice.

Does the avatar appear on every widget?

Only voice widgets. If the widget does not support voice calls, there is no avatar to show.

Getting Started

Step 1: Open the Agent

1. Log in to the VoiceInfra dashboard

2. Open an assistant or workflow

3. Go to Model & Voice

Step 2: Enable Video Avatar

1. Scroll to the Video avatar section

2. Turn the switch on

3. The picker opens if no avatar has been chosen

Step 3: Choose and Preview

1. Search or filter the catalog

2. Select an avatar

3. Review the preview beside your voice settings

Step 4: Tune and Publish

1. Turn expressions on if the avatar supports them

2. Choose the expressions to show

3. Save or publish the agent

Step 5: Test the Call

1. Start a portal test call

2. Confirm the avatar appears and moves with the voice

3. Open the website widget and start a voice call

4. Minimize the stage and confirm the call continues

Conclusion: Key Benefits of VoiceInfra Video Avatars

✅ Web-Only by Design

• Video where there is a screen

• Phone calls stay audio-only

• No wasted rendering on calls that cannot show it

✅ One Agent, Every Surface

• Same prompt, tools, knowledge, and routing

• Assistant or workflow support

• No separate video product

✅ A Catalog, Not a Single Face

• Realistic, illustrated, and 3D animated styles

• Search and filters

• Preview before publishing

✅ Expressions That Match the Voice

• Word-timed movement with supported voices

• Provider-driven expression controls

• Clear notes when a voice updates once per reply

✅ A Widget Built for Video

• Full-screen call stage

• Minimize to a live thumbnail

• Typed input and cards inside the call

• Still image until the first frame

✅ Beta With Clear Limits

• Labeled Beta in the product

• Provider-dependent catalog

• No surprise claims about phone video

Ready to put a face on your web calls? Turn on Video Avatars from Model & Voice.

👉 Open the Portal — Turn on Video avatar

👉 Schedule a Demo — See an avatar call live

Related Resources

Voice Agent Guides:

• How to Connect a Phone Number to Your AI Voice Agent

• Buy US Phone Numbers for AI Voice Agents

Avatar Providers:

• Anam — real-time interactive avatars

• Keyframe Labs — lifelike interactive avatars

Additional Reading:

• Website Widget Setup Guide

• Cold Transfer vs Warm Transfer vs Attended Warm Transfer