AvatarTalk AI

Generate talking-avatar videos from text in 17 languages and drop a live avatar agent into a LiveKit call with AvatarTalk. Agents produce personalized video responses without a studio or a recording session.

10 actions API key available

AvatarTalk AI website ↗ Platform Documentation ↗ Capability data verified 2026-08-01

An escalated support case needs a spoken answer in the customer's own language

Agent drafts the message text and resolves a character with List Avatars

Agent calls Request Video Invoice From Text to price the render before committing to it

Agent reads the estimated duration and the satoshi amount and compares them against the per-response ceiling

Agent posts the script, the runtime and the invoice amount into the review channel

The support lead approves payment, the agent polls Get Payment until it settles, then calls Generate Paid Video

What This Integration Enables

Two things here do not exist elsewhere in the catalog. The first is a live avatar. Create LiveKit Session places an avatar agent into an existing LiveKit room so it can present or converse in real time during a call, using room, server and participant tokens minted by your own LiveKit deployment. That is a synthetic presenter joining a live conversation as a participant, not a video file being handed to someone afterward. The second is the billing model. Alongside ordinary account credits, AvatarTalk supports a pay-per-video path built on Bitcoin Lightning: Request Video Invoice From Text prices the render and returns a BOLT11 invoice with the estimated duration and the amount in satoshis, and Generate Paid Video renders only once that invoice has settled.

The rest is the recorded path, and it is deliberately simple. Generate Avatar Video is synchronous. It returns hosted MP4 and browser viewer URLs together with the duration, file size and credits consumed, so the links pass straight into the next action without any file handling. Delayed Generation is a genuinely useful variant: it returns the URLs immediately and defers both the render and the charge until someone first opens the link, which is the right setting for an email or landing page where most recipients will never press play. Billing runs at one second of account video time per second of output, and live sessions bill for as long as they stay connected, which is why Delete LiveKit Session is a required step rather than good hygiene. Metered spend and a persuasive synthetic presenter are exactly the conditions human-in-the-loop exists for.

Without FlowRunner

Recording is a scheduling problem A spoken response needs a person, a camera and a calendar slot before it can exist

Spend discovered after the fact Render costs land on a monthly invoice with no per-item decision behind any of them

Live sessions left running An avatar joined to a call keeps consuming video time after everyone has hung up

With FlowRunner

Speech as a workflow step A response is voiced in one of 17 languages inside the flow that decided what to say

Priced before it is paid The cost of a specific render is known and approved before the render is authorized

Sessions owned end to end The flow that opens a live session is the flow that closes it

Use Case Scenarios

Spoken responses on an escalated support case

A support case is escalated with a resolution that is hard to follow in writing: a configuration change with four steps and a screenshot. The agent drafts the explanation, picks a character with List Avatars, and calls Generate Avatar Video in the customer's language. The hosted MP4 URL is attached to the ticket in Zendesk and the reply goes out with the video embedded. Because Generate Avatar Video is synchronous and returns the URL along with the duration and the credits consumed, the flow knows immediately what the response cost and how long it runs, and can hold anything over the team's length limit for review.

A live avatar host in a scheduled briefing

A recurring product briefing runs in a LiveKit room. The agent calls Create LiveKit Session with the room name, the server URL and the tokens the deployment minted, and the avatar joins as a participant to deliver the opening segment. The flow calls Get LiveKit Session Status to confirm the agent actually joined before the human host begins, rather than assuming it. When the call ends, the same flow calls Delete LiveKit Session. That closing step is not optional. The session consumes video time for as long as it stays connected, so the flow that opened it owns the responsibility for shutting it down.

Reconciling pay-per-video spend

A team running the Lightning path bills each render individually rather than drawing on a pool. Requesting an invoice does not settle it, and an invoice that is issued but never paid produces no video and no charge. The agent runs a reconciliation pass with List Payments, which returns every payment record with its amount in satoshis and dollars, its paid or pending status and its timestamps. Invoices that have sat pending past the flow's window are reported so someone can decide whether the request should be reissued or dropped, and the totals go into the monthly spend record alongside every other metered service.

Human-in-Loop Highlight

The gate on this connector sits before the money moves, because on the Lightning path the money is the irreversible part. Request Video Invoice From Text returns a BOLT11 invoice with an amount in satoshis, and Generate Paid Video rejects an unpaid invoice outright, which means payment has to clear before anything is rendered. There is no preview to judge first, no chargeback window afterward, and no dispute path once a Lightning payment settles. The agent therefore does everything up to the invoice and then stops: "Support case 8812. Script is 41 seconds, invoice is 1,180 sats. This settles before the video exists and cannot be reversed. Approve payment, edit the script, or cancel?" A person answers, the agent polls Get Payment until the invoice reads settled, and only then calls Generate Paid Video. The same discipline covers the live path from the other direction. A LiveKit session bills for every second it stays connected, so the flow that opens one is written with its Delete LiveKit Session already in place, and an agent that cannot confirm the session closed raises it rather than moving on.

Agent processes routinely

Detects exception requiring judgment

Clear match Continues automatically

Ambiguous Routes to human via preferred channel

Human decides

Agent resumes with decision

Agent Capabilities

10 actions

Video Generation

Generate Avatar Video Renders a talking-avatar video that speaks the supplied text, with speech synthesized in the chosen language and emotional delivery. The response carries hosted MP4 and browser viewer URLs along with the video duration, file size and credits consumed, so the URLs pass straight to another action. Delayed Generation defers both the work and the credit charge until someone first opens the returned URL, which suits emails and landing pages where the video may never be watched. One second of generated video consumes one second of account video time.
List Avatars Returns the avatar characters currently available to the account, straight from AvatarTalk. This is how a flow discovers characters added after the extension shipped. The avatar dropdowns on the generation and session actions cover the fourteen documented characters; use this action when the identifier you need is not among them.

Live Sessions

Create LiveKit Session Places a live avatar agent into an existing LiveKit room so it can present or converse in real time during a video call. Supply the room name, the LiveKit server URL and the participant, listener and agent tokens minted by your own deployment. Returns a task ID. Sessions consume video time for as long as they stay active. Choosing the Expressive emotion falls back to Neutral for the avatar itself.
Get LiveKit Session Status Reports the current state of a live session by its task ID, which lets a flow confirm the agent actually joined the room before continuing, or detect a session that has already ended.
Delete LiveKit Session Terminates an active session and removes the agent from the LiveKit room. Because sessions bill while they are running, call this as soon as the call ends rather than leaving the agent connected.

Pay Per Video

Request Video Invoice From Text Prices an avatar video from the text it will speak and returns a Bitcoin Lightning BOLT11 invoice covering it, with the estimated duration and the amount in satoshis. The first step of the pay-per-video flow, which bills each render individually instead of drawing on account credits.
Request Video Invoice From Audio Prices an avatar video from an existing speech recording and returns a BOLT11 invoice covering it. The audio is fetched from the URL you supply and forwarded as a multipart upload, and must be a WAV file at 16 kHz, mono, 16-bit. Used when the narration already exists and the avatar only needs to lip-sync to it.
Generate Paid Video Renders the avatar video that a paid Lightning invoice covers and stores it in FlowRunner file storage, returning a downloadable URL. The invoice must already be settled; an unpaid invoice is rejected. AvatarTalk streams raw MP4 bytes rather than a link, so the extension validates the MP4 signature before uploading. Choosing the Expressive emotion falls back to Neutral for the avatar.
Get Payment Checks whether a Lightning invoice has settled, returning the amount in both satoshis and US dollars with the creation and payment timestamps. Poll this between requesting an invoice and calling Generate Paid Video so a render is only attempted once payment has cleared.
List Payments Returns every Lightning payment record on the account with its amount in satoshis and dollars, its paid or pending status and its timestamps. Used to reconcile spend on the pay-per-video path, or to spot invoices that were issued and never settled.

Frequently Asked Questions

What can FlowRunner do with AvatarTalk AI?

FlowRunner agents can run Generate Avatar Video, List Avatars, and Create LiveKit Session in AvatarTalk AI, plus 7 more actions.

Does connecting AvatarTalk AI to FlowRunner require OAuth?

No. AvatarTalk AI connects to FlowRunner with an API key, no OAuth flow required.

Can AvatarTalk AI trigger a FlowRunner workflow automatically?

AvatarTalk AI doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.

Start building with AvatarTalk AI

$100 in credits. No card required. Connect in minutes.

Start Building Free Talk to us