Give your Cartesia AI voice agent a face

A Cartesia AI voice agent already has the speech layer: Sonic-3.5 for expressive text-to-speech, Ink-2 for speech recognition, and Line for agent orchestration. Anam adds the face layer, turning that voice into a live video agent that can greet, sell, coach, train, or guide users on screen.

That visual layer is not cosmetic. In customer-reported testing shared with Anam, 70% of users preferred video over voice-only interaction. When a voice agent has a visible face, teams have a better shot at lifting engagement, sales conversion, and customer retention because the interaction feels closer to a human conversation.

Cartesia included this pattern as workflow #10 in its guide to 10 enterprise voice workflows you can build with Cartesia: "Voice for Digital Avatars." Cartesia's framing is simple: give your Cartesia voice agents a face, pair Sonic-3.5's low-latency speech with an on-screen avatar, and use it for demos, training, kiosks, and virtual reception.

Watch the Cartesia + Anam demo

Before we get into the build, here is the short demo recorded for Cartesia's launch. It shows the end state of workflow #10: a Cartesia voice agent speaking through an Anam avatar, so the user sees a face on screen instead of hearing a voice-only agent.

How does a Cartesia AI voice agent get a face?

The architecture has one extra stage after speech generation.

Cartesia produces the voice. Anam receives the audio and generates a lip-synced video stream of the avatar speaking those words. The user sees a real-time face rather than an audio-only call or a waveform.

The practical pipeline looks like this:

1. Your Cartesia agent decides what to say.

2. Sonic-3.5 turns that response into speech.

3. Anam's audio passthrough receives the PCM audio stream.

4. Anam renders the avatar video and streams it to your app.

This is the same broad pattern behind Anam's guide to building an AI voice agent with a face, and the same product idea behind the ElevenLabs voice agent avatar and Mistral voice agent avatar posts. The voice provider can change. The face layer stays the same.

For teams already building with Cartesia, that matters because you do not need to throw away the voice stack. You can keep the Cartesia pieces you chose and add interactive avatars where the user experience needs visual presence.

What does the video layer change for users?

Voice-only agents are efficient, but they can feel empty in high-trust workflows. There is no eye contact. No visible turn-taking. No expression that tells the user the agent is listening.

A face changes the interface. It gives the user something to orient around, which is especially useful when the workflow asks for attention, trust, or follow-through.

That is why Cartesia's #10 workflow calls out presenters for demos, training, kiosks, and virtual reception. Those are all moments where a voice can do the task, but a face can make the interaction easier to stay with.

The same applies to sales and customer success. A voice agent can qualify a lead or answer a support question. A video agent can do that while feeling more present. For related examples, see Anam's posts on AI avatars in sales enablement and conversational AI solutions.

Where does this fit in Cartesia's workflow guide?

Cartesia's guide covers ten enterprise voice workflows. The first three use Sonic-3.5 and Ink-2 directly through the API. Workflows four through nine use Line, Cartesia's voice agent framework, for phone-style agent flows such as support triage, appointment booking, KYC follow-up, role-play, sales qualification, and HR helpdesk.

Workflow ten is the visual one: voice for digital avatars.

That is the right split. Cartesia handles the voice infrastructure. Anam handles the face generation, video stream, and avatar experience. Together, the stack gives teams a way to build voice agents that users can see.

If you want to follow Cartesia's examples, the full demo code is in the Cartesia enterprise use cases repo, and Cartesia's Sonic-3.5 docs cover the TTS model itself.

How do you build it with Anam docs?

Start with Anam's documentation overview. It explains the core model: a persona combines a face, voice, LLM, and system prompt, and every live conversation ends with real-time interactive avatars turning speech audio into video.

For a Cartesia integration, the most relevant Anam path is Custom TTS with audio passthrough. That doc shows how to pipe speech audio from an external TTS provider into an avatar for real-time lip sync. The example uses ElevenLabs, but the docs explicitly describe the same pattern for any TTS provider, including Cartesia.

The build checklist is:

1. Create or choose the Anam avatar you want on screen.

2. Create an Anam session token with audio passthrough enabled.

3. Start the avatar stream in your web app with the JavaScript SDK.

4. Send Cartesia Sonic-3.5 audio chunks into Anam's agent audio input stream.

5. End the sequence when Cartesia finishes a response, and interrupt it when the user barges in.

Use the JavaScript SDK quickstart for the basic browser setup, voice configuration docs for persona voice setup, and the session token API reference for server-side session creation.

The main production detail is audio format. Anam's audio passthrough expects PCM16 at 16kHz mono. Your Cartesia stream should be converted or configured to match before you send chunks into Anam.

When should you add Anam to a Cartesia agent?

Add a face when the workflow depends on attention, trust, or completion.

Good fits include:

1. Product demos where the agent presents and answers questions.

2. Training and coaching flows where the learner needs to stay engaged.

3. Sales qualification where trust affects conversion.

4. Kiosks and reception flows where a visible agent feels more natural.

5. Customer retention moments where tone and presence matter.

For short, purely transactional flows, voice-only may be enough. A shipping status lookup or one-field confirmation might not need video. The video layer earns its keep when the experience needs to feel human enough for users to keep going.

That is the simplest way to evaluate the stack: Cartesia gives the agent a high-quality voice. Anam gives it a face. Use both when the business outcome depends on the user staying in the conversation.

Frequently asked questions

What is an AI voice agent?

An AI voice agent is software that listens to speech, decides how to respond, and speaks back using generated audio. In a Cartesia stack, Sonic-3.5 can provide text-to-speech, Ink-2 can provide speech recognition, and Line can orchestrate agent behavior.

Can a Cartesia voice agent use a video avatar?

Yes. A Cartesia voice agent can use a video avatar by sending its speech audio into a face generation layer such as Anam's audio passthrough mode.

Why add video instead of staying voice-only?

Video adds visual presence, which can help users stay engaged and trust the interaction. Anam has seen customer-reported testing where 70% of users preferred video over voice-only experiences.

How does Anam work with Cartesia Sonic-3.5?

Cartesia Sonic-3.5 generates the speech audio, and Anam uses that audio to render a lip-synced avatar video stream. The two systems can be connected through Anam's custom TTS audio passthrough flow.

Does Anam replace Cartesia?

No. Anam adds the face and video layer while Cartesia can continue to provide voice models and agent infrastructure.

Where should developers start?

Developers should start with Anam's custom TTS audio passthrough docs and the JavaScript SDK quickstart. Those docs cover session setup, browser streaming, and how to send external speech audio into an avatar.