Skip to content
Model infoModelsSnapshot: 15 Oct 2026

HeyGen / HeyGen real-time avatars

LiveAvatar

Managed real-time avatar renderer with image- or video-based identities, streamed speech input and WebRTC video delivery. FULL mode adds a separate managed conversation pipeline.

availableClosed-source managed componentVideo generationAudio-driven video
Official documentation (opens in a new tab)Reviewed 14 Oct 2026
11 of 24 specification fields collected
Output resolutionPreset avatars: 1080p library, Essential included custom avatar: 720p, Business included custom avatar: 1080pSee specification notes below
Clip durationFree: 2 minutes; Essential: 20 minutes; Business: 60 minutes per session; Enterprise customSee specification notes below
Input billingSee ratesLiveAvatar · Avatar Only
Output billingSee ratesPublished units and conditions below

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Maximum output
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Context caching
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Reasoning
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Tool calling
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Structured output
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Streaming
SupportedSource for Streaming (opens in a new tab)

Actual generated video streamed to a WebRTC room as supplied audio arrives.

Batch processing
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Fine-tuning
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Parameters
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Architecture
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Knowledge cutoff
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Languages
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Release date
Not collected

Underlying model/checkpoint details have not been established; missing information is not unsupported.

Input modalities
Image, Video, AudioSource for Input modalities (opens in a new tab)

Image or video creates the reference identity; streamed response audio drives the LITE renderer. Text belongs to upstream conversation/TTS components.

Output modalities
VideoSource for Output modalities (opens in a new tab)

Supplied audio accompanies the stream; generated speech is handled by a separate TTS or speech-to-speech component.

Sample rate
24,000Source for Sample rate (opens in a new tab)

Input speech: PCM 16-bit, 24 kHz, mono. This is not a generated-audio output sample-rate claim.

Maximum request body
1 MB per WebSocket JSON packetSource for Maximum request body (opens in a new tab)

Includes the JSON envelope and base64 audio. Recommended audio chunk approximately one second; not a session context limit.

Output formats
H264 video over WebRTCSource for Output formats (opens in a new tab)

VP8 remains in the schema but is explicitly deprecated.

Output resolutions
Preset avatars: 1080p library, Essential included custom avatar: 720p, Business included custom avatar: 1080pSource for Output resolutions (opens in a new tab)

Account, avatar and requested quality govern. FAQ warns 1080p can increase latency and requires compatible source footage.

Output duration
Free: 2 minutes; Essential: 20 minutes; Business: 60 minutes per session; Enterprise customSource for Output duration (opens in a new tab)

max_session_duration cannot exceed the subscription limit.

Provider-reported latency
Under 300 ms median time to first frameSource for Provider-reported latency (opens in a new tab)

Provider service-level marketing claim, not end-to-end conversational response time or a guarantee for every custom 1080p avatar.

Generation controls
Speak, end utterance, interrupt, listening/idle state and keep-alive commandsSource for Generation controls (opens in a new tab)

No arbitrary dynamic gesture control. The FAQ ties gestures to avatar training footage.

Usage metering
LITE: 1 credit/min; FULL: 2 credits/min. Credits and subscriptions are separate from HeyGen video API plans.Source for Usage metering (opens in a new tab)

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

LiveAvatar · Avatar Only

Provider source (opens in a new tab)
API model ID
Not collected

mode=LITE is an integration mode, and avatar_id selects an identity asset; neither identifies a model checkpoint.

Access
POST https://api.liveavatar.com/v1/sessions/token with mode=LITE, followed by POST /v1/sessions/start. Supply speech through the returned WebSocket; video streams into the WebRTC room.
Regions
No specific processing region or data-residency guarantee established for this model.
subscriptions
Free: 10 credits/month. Essential: $99/month, 1,100 credits and one 720p custom avatar. Business: $475/month, 6,000 credits and one 1080p custom avatar.Source for subscriptions (opens in a new tab)

Included credits are allowances, not a standalone per-minute tariff.

Concurrent requests
Free: one concurrent session; Essential and Business advertise unlimited concurrency. Enterprise capacity and SLA are contractual.Source for Concurrent requests (opens in a new tab)

No additional per-concurrency charge does not imply free usage or an unlimited session duration.

enterprisePricing
Advertised as low as $0.01 per minute at the top Enterprise volume tier; custom agreement required.Source for enterprisePricing (opens in a new tab)

Not a universal self-serve price.

avatarCreation
Video identity: continuous two-minute footage and consent recording; voice clone created from footage. Image identity: one photo, no automatic voice clone. Custom avatars consume slots; stock avatars do not.Source for avatarCreation (opens in a new tab)
recordingGuidance
Core guide: 15 seconds listening, 90 seconds speaking, 15 seconds idling. Intro help article describes the final 15 seconds as listening.Source for recordingGuidance (opens in a new tab)
quality
quality enum: very_high, high (default), medium, low. No exact resolution mapping for every quality value verified.Source for quality (opens in a new tab)
gestureControl
No dynamic arbitrary gesture control; movements derive from source footage. Listening and idle state commands remain available.Source for gestureControl (opens in a new tab)
Billing scope
Overage must be enabled. Essential $0.095/credit; Business $0.09/credit. Charges under $100 are billed next cycle; above $100 are automatically settled into the next overage cycle.Source for Billing scope (opens in a new tab)
compatibility
LiveAvatar and HeyGen prerecorded avatars are not cross-compatible. Interactive Avatar was the earlier beta; no automatic migration of every existing asset is guaranteed.Source for compatibility (opens in a new tab)
sessionConfiguration
Create session token at /v1/sessions/token, then start with /v1/sessions/start. max_session_duration must fit the plan. Stored voice_agent is the current FULL configuration primitive; inline avatar_persona is deprecated.Source for sessionConfiguration (opens in a new tab)
billingMeter
1 credit per minuteSource for billingMeter (opens in a new tab)
Data retention
LiveAvatar FAQ says Avatar Only/LITE does not retain session data. This statement does not establish retention for saved avatar assets, account records or external speech/LLM providers.Source for Data retention (opens in a new tab)
audioInput
PCM 16-bit 24 kHz mono, base64-encoded audio inside WebSocket JSON frames; maximum packet 1 MB. Approximately one-second audio chunks recommended.Source for audioInput (opens in a new tab)
transport
Default LiveKit room; bring your own LiveKit or Agora configuration. Audio commands use WebSocket; generated video uses WebRTC. H264 default; VP8 deprecated.Source for transport (opens in a new tab)
utterances
agent.speak appends audio; agent.speak_end seals the utterance. agent.interrupt clears pending audio/video. source_event_id correlates an utterance to its first speak event.Source for utterances (opens in a new tab)
timeout
session.keep_alive refreshes activity beyond the five-minute idle timeout, subject to the maximum session duration. Wait for connected state before sending commands.Source for timeout (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 14 Oct 2026
CategoryUsageRateUnit
Session overageEssential LITE overage1$0.095minute of streaming
Session overageBusiness LITE overage2$0.09minute of streaming

Rate conditions

  1. Calculated from the published Essential overage rate of $0.095/credit × 1 credits/minute. Requires the paid plan, exhausted allowance and enabled overage. Bring-your-own STT/LLM/TTS providers can bill separately. Enterprise $0.01/min floor is not applied.
  2. Calculated from the published Business overage rate of $0.09/credit × 1 credits/minute. Requires the paid plan, exhausted allowance and enabled overage. Bring-your-own STT/LLM/TTS providers can bill separately. Enterprise $0.01/min floor is not applied.
Pricing source (opens in a new tab)

Data & enterprise terms

LiveAvatar FAQ says Avatar Only/LITE does not retain session data. This statement does not establish retention for saved avatar assets, account records or external speech/LLM providers.

Read the service policy (opens in a new tab)

LiveAvatar · Full pipeline

Provider source (opens in a new tab)
API model ID
Not collected

mode=FULL is an integration mode, and avatar_id selects an identity asset; neither identifies a model checkpoint.

Access
POST https://api.liveavatar.com/v1/sessions/token with mode=FULL, followed by POST /v1/sessions/start. LiveAvatar manages the conversation stack, including separate STT/LLM/TTS components.
Regions
No specific processing region or data-residency guarantee established for this model.
subscriptions
Free: 10 credits/month. Essential: $99/month, 1,100 credits and one 720p custom avatar. Business: $475/month, 6,000 credits and one 1080p custom avatar.Source for subscriptions (opens in a new tab)

Included credits are allowances, not a standalone per-minute tariff.

Concurrent requests
Free: one concurrent session; Essential and Business advertise unlimited concurrency. Enterprise capacity and SLA are contractual.Source for Concurrent requests (opens in a new tab)

No additional per-concurrency charge does not imply free usage or an unlimited session duration.

enterprisePricing
Advertised as low as $0.01 per minute at the top Enterprise volume tier; custom agreement required.Source for enterprisePricing (opens in a new tab)

Not a universal self-serve price.

avatarCreation
Video identity: continuous two-minute footage and consent recording; voice clone created from footage. Image identity: one photo, no automatic voice clone. Custom avatars consume slots; stock avatars do not.Source for avatarCreation (opens in a new tab)
recordingGuidance
Core guide: 15 seconds listening, 90 seconds speaking, 15 seconds idling. Intro help article describes the final 15 seconds as listening.Source for recordingGuidance (opens in a new tab)
quality
quality enum: very_high, high (default), medium, low. No exact resolution mapping for every quality value verified.Source for quality (opens in a new tab)
gestureControl
No dynamic arbitrary gesture control; movements derive from source footage. Listening and idle state commands remain available.Source for gestureControl (opens in a new tab)
Billing scope
Overage must be enabled. Essential $0.095/credit; Business $0.09/credit. Charges under $100 are billed next cycle; above $100 are automatically settled into the next overage cycle.Source for Billing scope (opens in a new tab)
compatibility
LiveAvatar and HeyGen prerecorded avatars are not cross-compatible. Interactive Avatar was the earlier beta; no automatic migration of every existing asset is guaranteed.Source for compatibility (opens in a new tab)
sessionConfiguration
Create session token at /v1/sessions/token, then start with /v1/sessions/start. max_session_duration must fit the plan. Stored voice_agent is the current FULL configuration primitive; inline avatar_persona is deprecated.Source for sessionConfiguration (opens in a new tab)
billingMeter
2 credits per minuteSource for billingMeter (opens in a new tab)
Data retention
LiveAvatar FAQ says FULL retains session transcripts only; users may delete them through the session-events API. This does not establish the retention policies of external providers or saved avatar assets.Source for Data retention (opens in a new tab)
pipeline
VAD, STT, LLM, TTS and avatar video are separate pipeline components. Custom LLM and TTS integration supported; no fixed GPT-4o-mini default is inferred from an older FAQ answer.Source for pipeline (opens in a new tab)
providerSettings
Current session schema exposes provider-specific ElevenLabs, Fish Audio and Cartesia voice settings. These provider models retain their own identities and limits.Source for providerSettings (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 14 Oct 2026
CategoryUsageRateUnit
Session overageEssential FULL overage1$0.19minute of streaming
Session overageBusiness FULL overage2$0.18minute of streaming

Rate conditions

  1. Calculated from the published Essential overage rate of $0.095/credit × 2 credits/minute. Requires the paid plan, exhausted allowance and enabled overage. Full pipeline bundle, not an additive charge for every component. Enterprise $0.01/min floor is not applied.
  2. Calculated from the published Business overage rate of $0.09/credit × 2 credits/minute. Requires the paid plan, exhausted allowance and enabled overage. Full pipeline bundle, not an additive charge for every component. Enterprise $0.01/min floor is not applied.
Pricing source (opens in a new tab)

Data & enterprise terms

LiveAvatar FAQ says FULL retains session transcripts only; users may delete them through the session-events API. This does not establish the retention policies of external providers or saved avatar assets.

Read the service policy (opens in a new tab)

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

LiveAvatar · Avatar Only

Not verified

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

How it works
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

Minimum reusable input
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

Lifetime & refresh
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

How to enable it
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

Usage fields
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

Billing rules
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

Limits & data boundaries
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 14 Oct 2026

LiveAvatar · Full pipeline

Not verified

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

How it works
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

Minimum reusable input
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

Lifetime & refresh
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

How to enable it
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

Usage fields
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

Billing rules
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

Limits & data boundaries
Not collected

No model prompt-cache contract established. Session continuity, saved avatars and retained transcripts are separate.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 14 Oct 2026

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

HeyGen identifies LiveAvatar as its production real-time rendering technology and successor to Interactive Avatar. The hosted visual model is not exposed as a versioned checkpoint and its underlying weight identity is not disclosed, so this is explicitly a managed component, not a newly claimed foundation model.

Classification evidence (opens in a new tab)

Lifecycle dates

Provider describes Interactive Avatar transition to LiveAvatar6 Nov 2025Source (opens in a new tab)

Sources reviewed