Models

The models behind the HeyGen API: HeyGen Video for prompt-to-video, the Avatar V, Avatar IV, and Avatar III rendering engines, and HeyGen Voice for voice cloning and speech.

HeyGen Video

Prompt to video. Generates the whole frame.

Avatar V

Highest fidelity. Opt-in per look.

Avatar IV

Default engine. Broadest coverage.

Avatar III

Photo-to-video pipeline. 4K.

HeyGen Voice

Voice clones and speech.

The v3 API exposes three model families.

HeyGen Video

HeyGen Video generates video from a prompt, inventing the subject, the setting and the sound together in one call to POST /v3/models/videos.

HeyGen Avatar

Avatar rendering engines turn a look and a script into video, selected per request through the engine field on POST /v3/videos. New integrations should default to Avatar IV for broad coverage, opt into Avatar V for the highest-fidelity motion and lip-sync, and use Avatar III for its dedicated photo-to-video pipeline.

{
  "type": "avatar",
  "avatar_id": "YOUR_LOOK_ID",
  "script": "Hello from HeyGen.",
  "voice_id": "YOUR_VOICE_ID",
  "engine": { "type": "avatar_iv" }
}

Before requesting an engine, confirm the look supports it: supported_api_engines on GET /v3/avatars/looks/{look_id} lists which of avatar_iii, avatar_iv, and avatar_v the look accepts. Requesting an engine that isn’t listed returns 400.

	Avatar V	Avatar IV	Avatar III
engine.type	avatar_v	avatar_iv	avatar_iii
API version	v3	v3	v3
Default engine	✗ (explicit opt-in)	✓	✗ (explicit opt-in)
Available to new users	✓	✓	✓
Digital Twin	✓	✓	✓
Photo Avatar	✗	✓	✓ (dedicated pipeline)
Studio Avatar	✗	✓	✓ (renders as Digital Twin)
Arbitrary image input	✗	✓	✗
motion_prompt	✓	✓	✗
expressiveness	✗	✓	✗
Cross-reference animation	✓	✗	✗
Eligibility check required	✓	✗	✗

A separate, older Avatar III engine remains available to existing customers through the legacy v1/v2 endpoints. It uses a different pipeline and is not the same as the Avatar III (avatar_iii) engine documented here. See Avatar III for the distinction.

HeyGen Voice

HeyGen Voice is HeyGen’s in-house voice model. Clone a speaker in one of two modes on POST /v3/models/audio/voices; both speak through Text to Speech, completed or streamed.

	Instant Clone	Professional Clone
mode	instant	professional
Source audio	One recording	1–10 recordings, 20+ minutes in total
Ready in	Seconds	Asynchronous training; poll until ACTIVE
Availability	Beta. Free to create during the preview	Beta. Paid: one purchased voice clone slot per voice

Stock and designed voices from the voice catalog speak through Third Party Speech instead.