live
MODEL IDheygen:avatar@6

HeyGen Avatar 6

HeyGen
by
Available with Zero Data Retention

HeyGen Avatar 6 is an audio-driven avatar video model that animates a HeyGen avatar, a single photo, or a reference video of an animated or illustrated character. It generates gestures and facial expressions that match what is being said, with accurate lip sync and native 2K output, and a motion prompt can direct gestures, posture and camera. It suits presenter, training and product videos where natural motion matters.

HeyGen Avatar 6

Animating a photo

How to animate a single photo with HeyGen Avatar 6: framing and output shape, illustrated and 3D characters, resolution and long scripts.

Introduction

One photo and a speech track are enough for HeyGen Avatar 6 to animate the person in the photo, with lip sync and gestures that follow the words. The photo stays the shot: its framing, setting and light carry into the video, and the video runs as long as the audio.

This is the mode for turning an existing photo into a presenter: a real estate agent in the listing they are selling, a creator holding the product, a support agent answering a ticket, a mascot greeting a child in an app.

The agent, the room and the window light all come from the photo
{
  "taskType": "videoInference",
  "taskUUID": "756118ea-1a3a-4ce4-9406-d1f4a539f359",
  "model": "heygen:avatar@6",
  "inputs": {
    "referenceImages": ["https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-realtor.jpg"],
    "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-realtor.mp3"
  }
}
Built from 2 references
  • Photo
  • Speech
    0:00

    This two-bedroom gets morning sun in every room. The kitchen was redone last spring, the floors are original oak, and there's a shared roof terrace with a view you'll want to see for yourself.

This guide covers the request, how the photo sets the framing and the shape of the output, illustrated and 3D characters, resolution, and long scripts.

The request

The photo goes in inputs.referenceImages and the speech in inputs.audio.

Try in Playground
import { createClient } from '@runware/sdk'

const client = await createClient({ apiKey: process.env.RUNWARE_API_KEY })
await client.connect()

const [result] = await client.run({
  model: 'heygen:avatar@6',
  inputs: {
    referenceImages: [
      'https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-realtor.jpg'
    ],
    audio: 'https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-realtor.mp3'
  }
})
import asyncio
import os

from runware import Runware


async def main():
    async with Runware(api_key=os.environ["RUNWARE_API_KEY"]) as client:
        results = await client.run({
            "model": "heygen:avatar@6",
            "inputs": {
                "referenceImages": [
                    "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-realtor.jpg"
                ],
                "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-realtor.mp3"
            }
        })


asyncio.run(main())
curl https://api.runware.ai/v1 \
  -H "Authorization: Bearer $RUNWARE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '[
    {
      "taskType": "videoInference",
      "taskUUID": "756118ea-1a3a-4ce4-9406-d1f4a539f359",
      "model": "heygen:avatar@6",
      "inputs": {
        "referenceImages": [
          "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-realtor.jpg"
        ],
        "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-realtor.mp3"
      }
    }
  ]'
runware run heygen:avatar@6 \
  inputs.referenceImages.0=https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-realtor.jpg \
  inputs.audio=https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-realtor.mp3
{
  "taskType": "videoInference",
  "taskUUID": "756118ea-1a3a-4ce4-9406-d1f4a539f359",
  "model": "heygen:avatar@6",
  "inputs": {
    "referenceImages": [
      "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-realtor.jpg"
    ],
    "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-realtor.mp3"
  }
}
Response
[
  {
    "taskType": "videoInference",
    "taskUUID": "756118ea-1a3a-4ce4-9406-d1f4a539f359",
    "videoUUID": "ce94a82a-c2e5-4347-ab83-f03ad9ebf139",
    "videoURL": "https://vm.runware.ai/video/os/a14d18/ws/2/vi/ce94a82a-c2e5-4347-ab83-f03ad9ebf139.mp4"
  }
]
  • inputs.referenceImages takes one image of up to 32 MB in JPEG, PNG or WebP, as a UUID (a previous generation's, or one uploaded via Media Storage), a URL, a data URI or base64.
  • inputs.audio is required with a photo and is the only source of speech in this mode, so a speech script is rejected. To speak a script, generate the audio first with HeyGen Voice 1.0.
  • The video runs the length of the audio. There is no duration to set.
  • width and height are rejected with a photo, because the output keeps the photo's aspect ratio. resolution sets its size, covered in Resolution.
  • inputs.background, settings.backgroundColor, settings.removeBackground, settings.caption and settings.fit work with a HeyGen avatar in inputs.avatar and are rejected with a photo. Compose the background in the photo itself.
  • positivePrompt directs gestures, expression, posture and camera, covered in the prompting guide. The examples here send none, so every gesture is the model's own.

Framing and output shape

Avatar 6 keeps the photo's framing and shape. Choose the photo for where the video will play, because it decides the output before the model does anything else.

A vertical photo held at arm's length gives a vertical video for Reels and TikTok:

A 9:16 photo gives a 9:16 video
{
  "taskType": "videoInference",
  "taskUUID": "7cfed11a-f79b-4c9d-9e7f-1786acf6dcfd",
  "model": "heygen:avatar@6",
  "inputs": {
    "referenceImages": ["https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-creator.jpg"],
    "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-creator.mp3"
  }
}
Built from 2 references
  • Photo
  • Speech
    0:00

    Okay, so this is the serum I've been using every single night for a month. Two drops, pat it in, done. My skin has never felt this soft.

A full-length photo keeps the whole body in frame, which suits fitness and wellness content where the stance matters as much as the face:

A full-length photo stays full length
{
  "taskType": "videoInference",
  "taskUUID": "967d4b52-5e8f-4aef-88e8-319dc62ff49a",
  "model": "heygen:avatar@6",
  "inputs": {
    "referenceImages": ["https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-yoga.jpg"],
    "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-yoga.mp3"
  }
}
Built from 2 references
  • Photo
  • Speech
    0:00

    Welcome to your morning flow. We'll start standing, feet hip-width apart, and take a few easy breaths before we move.

A tight head-and-shoulders crop gives a close, conversational shot, the framing of a support or sales reply:

A square crop gives a square video
{
  "taskType": "videoInference",
  "taskUUID": "ee581dc5-7946-4e0f-b6ce-b4484a5fa92a",
  "model": "heygen:avatar@6",
  "inputs": {
    "referenceImages": ["https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-support.jpg"],
    "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-support.mp3"
  }
}
Built from 2 references
  • Photo
  • Speech
    0:00

    Thanks for reaching out. I've checked your account, and the refund went through this morning. You should see it on your card within three business days.

Frame the photo for the shot you want. A waist-up photo leaves room for the gestures Avatar 6 adds. A tight crop does not stop them: the support agent's hand comes up into the close-up to wave on the greeting and to count the three days. Cropping or padding the photo before you send it is the way to change the output's shape, since width and height are not available with a photo.

Illustrated and 3D characters

The reference can be an illustration or a 3D render as well as a photograph. Avatar 6 animates a drawn or rendered character the same way and keeps its art style, which suits app mascots and animated explainers.

A flat 2D mascot keeps its outlines and colors
{
  "taskType": "videoInference",
  "taskUUID": "68d33d98-3850-4712-a28d-501c5dfbf6ea",
  "model": "heygen:avatar@6",
  "inputs": {
    "referenceImages": ["https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-mascot.jpg"],
    "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-mascot-cat.mp3"
  }
}
Built from 2 references
  • Illustration
  • Speech
    0:00

    Hi, friend! Today we're learning the names of five fruits. Listen to each word, then say it back to me. Ready? Let's go!

A 3D character keeps its rendering and lighting
{
  "taskType": "videoInference",
  "taskUUID": "392ad955-f9d0-4a81-b3c2-45c6d48ca2ba",
  "model": "heygen:avatar@6",
  "inputs": {
    "referenceImages": ["https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-explorer.jpg"],
    "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-explorer-kid.mp3"
  }
}
Built from 2 references
  • 3D render
  • Speech
    0:00

    Look how tall these trees are! Some of them are more than two hundred years old. Come on, let's find out what lives at the very top.

Draw the character facing the viewer with its mouth in view, the way you would frame a person for this mode, since the mouth is what the lip sync animates. For a character that already exists as animation, a clip of it moving carries more than a still, which the reference videos guide covers.

Resolution

With a photo, resolution sets the size of the output and the photo sets the shape. 1080p is the default and 2k the larger option. The landscape photos in this guide, 2752 × 1536, come back at 1920 × 1072 with 1080p and at 2880 × 1608 with 2k. The vertical photo comes back at 1072 × 1920 and the square crop at 1088 × 1088.

resolution: 2k
{
  "taskType": "videoInference",
  "taskUUID": "9a969dc3-c34a-4c9a-8fe1-4b98c082cbd5",
  "model": "heygen:avatar@6",
  "inputs": {
    "referenceImages": ["https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-chef.jpg"],
    "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-chef.mp3"
  },
  "resolution": "2k"
}
Built from 2 references
  • Photo
  • Speech
    0:00

    This is our fresh tagliatelle with brown butter and sage. We make the pasta every morning, and it goes from the pot to your plate in under two minutes.

A frame three seconds into each render, with the loupe on the beard. Each render is its own take, so the pose differs:

Reach for 2k when the video will be cropped in an edit or played on a large screen, where the extra pixels in hair and skin survive the crop. For phone feeds and in-app players, 1080p is already at or above the display size.

Long scripts

A longer script is a longer audio file in the same request. The trainer below speaks for 32 seconds from one photo and one audio file, with nothing stitched together.

One request for the whole module intro
{
  "taskType": "videoInference",
  "taskUUID": "c9d375ff-c692-4a23-9400-ec75ce47b2f4",
  "model": "heygen:avatar@6",
  "inputs": {
    "referenceImages": ["https://assets.runware.ai/guides/heygen-avatar-6/reference-images/reference-trainer.jpg"],
    "audio": "https://assets.runware.ai/guides/heygen-avatar-6/reference-images/speech-trainer.mp3"
  }
}
Built from 2 references
  • Photo
  • Speech
    0:00

    Welcome to this short module on reporting a safety concern. Over the next few minutes, we'll cover what counts as a concern, who to tell, and what happens after you report it. Let's start with the first question. A safety concern is anything that could hurt you, a colleague or a visitor, even if nobody has been hurt yet. A loose cable across a walkway counts. So does a fire door that has been propped open, or a machine guard that is missing. If you are not sure whether something counts, report it anyway. Nobody will be in trouble for raising a concern that turns out to be minor.

Her face and framing hold from the first line to the last, so a course module or an onboarding video can be one request per section rather than a chain of short clips. Split a script only where the content changes, such as one video per lesson.

Tips

  1. One person, facing the camera, with the face clearly lit. The photo is everything the model knows about how the person looks.

  2. Shoot or crop for the destination. The output keeps the photo's shape, so a vertical photo is the way to a vertical video.

  3. Put the background in the photo. Background settings are rejected with a photo, and the room and the light in the photo are what the video shows.

  4. Leave room for the hands. A waist-up photo gives the gestures space. In a tight crop they still come up, close to the camera.

  5. Make the speech first. A script has to become audio before it reaches Avatar 6 in this mode, and HeyGen Voice 1.0 turns text into the speech track in one request.

  6. Render at 2K for footage you will crop. At display size the two resolutions look alike, and the difference is in the detail a crop or a large screen reveals.