Your chatbot already knows the answers. A real-time talking avatar lets it say them out loud, with a face, a voice, lip sync and natural gestures, so conversations feel closer to talking to a person than reading a text box. The good news is that you do not need to rebuild or replace your existing AI to do it.
This guide explains how to add a speaking avatar to an existing chatbot or AI agent, what to look for in an avatar layer, and how Captivation Inside works.
Do I need to replace my chatbot?
No. The most practical approach is to keep the AI you already have and add an avatar layer on top. Your chatbot, LLM or AI agent keeps generating the intelligence and the responses. The avatar layer turns each response into speech, lip sync, expressions and gestures, and shows it in your website or app.
That means you keep your existing prompts, data, agents, tools and workflows. Only the way the answer is delivered changes.
How adding a talking avatar works
With Captivation Inside, the process has three parts:
- Add the avatar widget to your front end. Captivation provides a JavaScript widget that displays the avatar on your website or app.
- Connect your AI to the Captivation Platform. Your LLM output is passed to Captivation through the Captivation Bridge.
- Captivation brings the response to life. The platform generates the voice and avatar instructions and delivers them to the browser in real time over a WebSocket connection. Voice, text, lip sync, gestures and expressions are synchronised as the conversation happens.
Captivation Inside is LLM agnostic, so it works with whichever model or chatbot platform you use. Full technical details are in the Captivation developer guide.
What to look for in an avatar layer
When you compare ways to add a talking avatar to your chatbot, check these points:
- Works with your existing AI. You should not have to move to a new chatbot platform or LLM.
- Response speed. Measure the time from your AI producing an answer to the avatar starting to speak. With Captivation this is typically 200 to 300 milliseconds when stored audio can be reused, and 600 to 800 milliseconds when new audio must be generated.
- Running cost per minute. Real-time avatars can be expensive to run at scale. Check the cost per minute of conversation, not just the monthly plan.
- Expressiveness. Lip sync alone is not enough. Gestures and expressions make the avatar feel engaged.
- Visual content. The best experiences show products, images, maps or video alongside the avatar, in sync with what it is saying.
- Device support. Visitors should not need to download an app. Captivation runs in modern web browsers on phones, tablets and PCs.
Why browser rendering matters
Most real-time avatar platforms continuously generate video on a server and stream it to the user. That needs a lot of compute and bandwidth for every minute of conversation.
Captivation works differently. Pre-created visual assets are delivered to the browser and brought to life with lightweight real-time instructions, so the avatar is rendered on the user's own device. This reduces compute and bandwidth and helps Captivation deliver faster connections, expressive gestures and costs that can be up to 90% lower than conventional real-time video avatar services.
Captivation's Voice Optimisation also matches around 80% of response phrases to audio that has already been generated for a voice, and reuses it instead of making a new text-to-speech request. New speech is still generated whenever it is needed.
How much does it cost?
Captivation Inside is usage-based. Avatar use costs less than one cent per minute of normal conversation, and a typical total cost for an avatar with voice is around 1 to 2 cents per minute, which is roughly $0.60 to $1.20 per hour. Plans start from $100 (£75) a month. Your own LLM costs are separate, because you keep your existing AI.
There is a free developer trial with credits for about 30 to 40 minutes of conversation, subject to availability.
Choosing an avatar and voice
You can choose from Captivation's growing range of avatars, voices and accents, including Jane, Aimi, Michael and Usain. Enterprise customers can commission bespoke branded avatars, characters and mascots. Captivation also supports compatible ElevenLabs voices for clients using their own ElevenLabs account, and multilingual experiences in more than 30 languages.
Frequently asked questions
Can I add a talking avatar to any chatbot? Captivation Inside is designed to work with existing chatbots, LLMs and AI agents. Your AI generates the response and Captivation turns it into a spoken, animated experience.
Does it work with voice agents such as Vapi or LiveKit? The technology can work with most voice-agent architectures, including services such as Vapi and LiveKit. The fully supported commercial offering for this use case is still being developed, so contact the team to discuss your requirements.
Do users need to install anything? No. The avatar runs in a normal web browser on smartphones, tablets and PCs.
How do I get started? Request the free Captivation Inside developer trial, or contact the team to discuss your integration. See pricing for plan details.
