HeyGen Avatar 6 on Segmind: talking-avatar video from a photo, a clip or a look
HeyGen Avatar 6 makes lip-synced talking-avatar videos with gestures that follow the words. Drive it from one photo, a short clip of an animated character, or a ready-made HeyGen presenter look, in a stock voice or your own audio.

HeyGen Avatar 6 is HeyGen's talking-avatar model, and it reads your script before it moves. It times pointing, counting and directional gestures to the matching words, keeps the motion fresh on every line, and lip-syncs to a HeyGen stock voice or your own audio. On Segmind it ships as three modalities. Image to video animates one photo. Reference to video makes an animated character talk from a short clip. Identity to video drives a ready-made HeyGen presenter look. All three render up to native 2K, and the clip runs as long as the speech.
Image to video
Give image to video one clear photo of a person or character and one source of speech, and it animates the face and upper body to match. Choose a script read by a HeyGen stock voice, or upload your own audio — never both. It reproduces the person's look from a single still, then performs the words instead of idly mouthing them. Reach for it to turn a founder, presenter or teacher headshot into a talking clip.
We shipped three things this month. First, the panels fold completely flat. Second, each one weighs under two kilos. And third — my favorite — they still charge in full shade.
Her fingers count one-two-three on "first, second, third" and she warms into a smile on the closing line. Write real beats — a number or a short list — so the gesture timing has something to track.
Here the speech comes from an uploaded WAV, MP3 or M4A, with no typed script or voice, while the motion prompt still directs the hands.
This is the refreshed identity. The wordmark keeps its weight, the spacing opens up a touch, and the palette now runs from ink to a single warm accent.
With motion_expressiveness set to less she delivers steadily — small nods, hands at rest, a composed face — instead of big gestures. Reach for it on briefings and other formal clips.
Use image to video from the API
One call takes the reference photo, the speech and the motion prompt.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"heygen-avatar-6-image-to-video",
reference_image="https://segmind-resources.s3.amazonaws.com/input/heygen-avatar-6-image-to-video-ref-f37b689c.jpg",
script="We shipped three things this month. First, the panels fold completely flat. Second, each one weighs under two kilos. And third — my favorite — they still charge in full shade.",
voice_id="4270665fc47f44aa989c093bb3dc2b99",
motion_prompt="Keep steady eye contact with the camera and use small, natural hand gestures; count on her fingers one, two, three as she says first, second, third; lean in slightly and smile warmly on the last line.",
tier="standard",
resolution="1080p",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/heygen-avatar-6-image-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"reference_image": "https://segmind-resources.s3.amazonaws.com/input/heygen-avatar-6-image-to-video-ref-f37b689c.jpg",
"script": "We shipped three things this month. First, the panels fold completely flat. Second, each one weighs under two kilos. And third — my favorite — they still charge in full shade.",
"voice_id": "4270665fc47f44aa989c093bb3dc2b99",
"motion_prompt": "Keep steady eye contact with the camera and use small, natural hand gestures; count on her fingers one, two, three as she says first, second, third; lean in slightly and smile warmly on the last line.",
"tier": "standard",
"resolution": "1080p"
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
-H "x-api-key: YOUR_API_KEY"
# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
-H "x-api-key: YOUR_API_KEY"Reference to video
Reference to video makes an animated or illustrated character talk. Supply a short reference clip of a cartoon, 3D, illustrated or mascot character, at least two seconds with the mouth visible. It learns the look, then lip-syncs it to a script or your own audio. Videos of real people are refused here, so use image to video for a person. Add an optional still of the same character to keep it on-model across renders.
Hello there, and welcome! Did you know a red fox makes over twenty different sounds? My three favourites are a yip, a bark, and a happy little gekker - listen closely!
A 3D cartoon fox lip-syncs the line and lands a paw-wave, a three-count and a paw cupped to its ear on cue. Write each gesture next to the exact words it should hit in the motion prompt.
No script is typed here; send a WAV, MP3 or M4A and leave the voice empty, and the clip runs as long as the audio.
The new wordmark keeps its weight, opens up the spacing a little, and still reads clearly even at a tiny size.
An illustrated cat keeps its paws clasped and speaks with only small nods and a steady expression. Pair "less" with a motion prompt that asks for minimal gestures.
Use reference to video from the API
The reference clip sets the character and framing; the script and voice set the speech.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"heygen-avatar-6-reference-to-video",
reference_video="https://segmind-resources.s3.amazonaws.com/input/heygen-avatar-6-reference-to-video-refvideo-5b29b86a.mp4",
reference_image="https://segmind-resources.s3.amazonaws.com/input/heygen-avatar-6-reference-to-video-refimage-a735e29f.jpg",
script="Hello there, and welcome! Did you know a red fox makes over twenty different sounds? My three favourites are a yip, a bark, and a happy little gekker - listen closely!",
voice_id="c30ecab014cb4a8d816d07f6b2b625c8",
motion_prompt="The fox waves a friendly open paw on 'Hello there, and welcome'. On 'three favourites' it holds up three claws and counts them off one by one. On 'listen closely' it leans in slightly and cups a paw near its ear. Warm, cheerful expression with small natural head movements; keep both paws visible.",
tier="standard",
resolution="1080p",
motion_expressiveness="default",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/heygen-avatar-6-reference-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"reference_video": "https://segmind-resources.s3.amazonaws.com/input/heygen-avatar-6-reference-to-video-refvideo-5b29b86a.mp4",
"reference_image": "https://segmind-resources.s3.amazonaws.com/input/heygen-avatar-6-reference-to-video-refimage-a735e29f.jpg",
"script": "Hello there, and welcome! Did you know a red fox makes over twenty different sounds? My three favourites are a yip, a bark, and a happy little gekker - listen closely!",
"voice_id": "c30ecab014cb4a8d816d07f6b2b625c8",
"motion_prompt": "The fox waves a friendly open paw on '\''Hello there, and welcome'\''. On '\''three favourites'\'' it holds up three claws and counts them off one by one. On '\''listen closely'\'' it leans in slightly and cups a paw near its ear. Warm, cheerful expression with small natural head movements; keep both paws visible.",
"tier": "standard",
"resolution": "1080p",
"motion_expressiveness": "default"
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
-H "x-api-key: YOUR_API_KEY"
# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
-H "x-api-key: YOUR_API_KEY"Identity to video
Identity to video needs no upload. Pick a HeyGen public avatar look by its ID, type a script, and the look performs it to camera. The look must be one that supports Avatar 6, and custom or cloned avatars are refused. Leave the voice empty to use the look's own, or set a HeyGen stock voice to change the delivery. Choose the frame you need, from 16:9 for courses to 9:16 for Shorts and Reels.
Here's how I'd lay out a small studio in three zones. First, the bed tucks into the far corner, away from the door. Second, the kitchen runs along this wall. And third, right under the window, the desk, where the daylight is best. Three zones, one room, no clutter.
The presenter counts the zones on a raised hand and points to each as she names it. Write numbered, directional lines and the gestures follow; keep scripts tight, since length sets the clip.
This card drops the script and lip-syncs supplied audio instead; send a WAV, MP3 or M4A under 50 MB, and the voice setting is ignored on this path.
Our refreshed identity keeps the wordmark's weight, opens the spacing a touch, and lets one warm accent carry the whole palette.
With motion_expressiveness set to less, the presenter sits still with hands clasped and only small nods. Pair it with a composed stock voice for corporate reads.
Use identity to video from the API
Name the look and the script, and pick the tier, resolution and frame.
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"heygen-avatar-6-identity-to-video",
avatar_id="51753ef612cc482cb46084c3165e1d20",
script="Here's how I'd lay out a small studio in three zones. First, the bed tucks into the far corner, away from the door. Second, the kitchen runs along this wall. And third, right under the window, the desk, where the daylight is best. Three zones, one room, no clutter.",
tier="standard",
resolution="1080p",
aspect_ratio="16:9",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/heygen-avatar-6-identity-to-video" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"avatar_id": "51753ef612cc482cb46084c3165e1d20",
"script": "Here'\''s how I'\''d lay out a small studio in three zones. First, the bed tucks into the far corner, away from the door. Second, the kitchen runs along this wall. And third, right under the window, the desk, where the daylight is best. Three zones, one room, no clutter.",
"tier": "standard",
"resolution": "1080p",
"aspect_ratio": "16:9"
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID/status" \
-H "x-api-key: YOUR_API_KEY"
# 3. Fetch the result
curl -s "https://api.segmind.com/v2/requests/$REQUEST_ID" \
-H "x-api-key: YOUR_API_KEY"Shared settings
| Setting | Value | Members | When to use it |
|---|---|---|---|
| Speech | A script with a HeyGen stock voice, or your own audio — never both | All three | Type a script, or upload WAV, MP3 or M4A to use a recorded voice |
| voice_id | A HeyGen stock voice | All three (optional on identity) | Required with a script on image and reference; optional on identity, which falls back to the look's own voice. Cloned voices are refused |
| tier | standard or turbo | All three | standard for full quality, turbo for faster drafts and batches |
| resolution | 1080p or 2k | All three | 1080p for social, 2k for large screens and fine detail |
| aspect_ratio | 16:9, 9:16, 1:1, 4:5, 5:4 or auto | Identity to video | Set the frame on identity; image and reference keep the input's shape |
| motion_prompt | Up to 1,200 characters | All three | Tie a gesture, expression or posture to a phrase |
| motion_expressiveness | default or less | All three | less for a calmer, steadier read |
| voice_speed | 0.5–2 on image and reference, 0.5–1.5 on identity | All three | Keep near 1 for a natural pace |
| language | A code like en or pt-BR, or auto-detect | Image, reference | Pin the script language, or leave empty to auto-detect |
Good to know
- Speech is a script with a HeyGen stock voice or your own audio, never both; a script always needs a stock voice, and cloned voices are refused on all three.
- Reference to video takes only animated or illustrated character clips — cartoon, 3D, illustrated or mascot — of at least two seconds; videos of real people return an error, so use image to video for a person.
- Identity to video accepts only HeyGen public avatar looks that support Avatar 6; your own custom or cloned avatars are refused.
- From a single photo, image to video gives a lower likeness than a reference clip would, and a sharp, front-facing face with the mouth visible gives the cleanest lip-sync.
- These are direct-address presenter models, not scene generators, so there is no b-roll, environment or multi-character action to prompt.
- The clip runs as long as the speech, so a long script means a longer render.
Resources
- Try HeyGen Avatar 6 Image to Video on Segmind · API reference
- Try HeyGen Avatar 6 Reference to Video on Segmind · API reference
- Try HeyGen Avatar 6 Identity to Video on Segmind · API reference
- HeyGen Avatar 6 model API documentation — inputs, the motion_prompt and motion_expressiveness controls, resolution and output sizing, and how to submit and retrieve a job
- HeyGen Avatar 6 model overview — what is new in Avatar 6: context-aware gestures, fresh motion every line, native 2K and higher realism
- HeyGen Avatar 6 Technical Report — the architecture, the realism evaluation, and the standard and turbo tiers
- #heygen-avatar-6
- #heygen
- #model-launch
- #video-generation
- #talking-avatar

