HeyGen Video: text, image or reference to video with native audio
Generate complete scenes with native dialogue, ambience and sound effects from a prompt, image or references — one HeyGen model with three modes.
HeyGen Video turns a single prompt into a finished scene — subject, setting, light and sound in one call. HeyGen built it on MiniMax H3 and post-trained it for contained production shots. Every clip ships with its own audio: dialogue, ambience and effects, with no separate speech pass or lip-sync step. One slug runs three modes, picked by what you send: a prompt alone, a prompt plus a first-frame image, or a prompt plus references.
ModelHeyGen VideoText, image or reference to video with native audio.What it does well
Whole scenes with their own audio
Send a prompt and you get the whole frame plus a matching soundtrack. In this run a cheesemonger speaks her exact quoted line in lip-sync, over a knife tap, a doorbell and low shop murmur. Picture and sound come from the one prompt, so the dialogue lands on her mouth and the ambience fits the room. Quote the line exactly — about 2.5 words a second — name the room tone, and add "no music" when you do not want a bed.
Keep a product or person in a new scene
Reference to video builds a fresh scene around something you supply and keeps it recognisable. Here a silk scarf from a reference image comes back with its terracotta-and-sage diamond print intact, knotted at the model's neck in a new studio shot. You can pass up to nine images, three videos and three audio files, addressed in the prompt as <Picture 1>, <Video 1> and <Audio 1>. Give one clean, well-lit reference and set an aspect ratio if you do not want the reference's own shape.
Start from your own first frame
In image to video, your still becomes the literal first frame and the clip animates on from there. This astronaut-and-lander photo opens the shot, then a gloved hand waves slowly while a radio-voiced line plays over a steady suit hiss. The image sets the aspect ratio, so crop it before you upload and leave aspect_ratio off. A first frame cannot be combined with references — pick one or the other.
Examples to try
One more text-to-video shot renders short lettering inside the frame.
A muralist in paint-spattered denim overalls steps back from a plain grey concrete wall freshly painted in tall, bold brush lettering with the single word BLOOM. The only lettering anywhere in frame is the word BLOOM, spelled exactly B L O O M. Shot on an ARRI Alexa Mini LF, 35mm T2, locked off on a tripod, no camera movement. She stands beside the finished wall with her hands at her sides, looks at it, then turns to camera and says: "Took me all morning." Flat daylight from the left. Natural skin texture, pores and fine lines, no retouching, no beauty filter, no glamour lighting. Muted colour, no teal-orange grade. No other logos, brand names, printed words or badges anywhere in frame. Sound: her voice close and clear, a distant street hum, one passing bicycle bell. No music.
The word BLOOM lands correctly spelled and is the only lettering in frame as the muralist turns and speaks. Spell the word out, name it as the only lettering, and keep it short — paragraphs and small print are unreliable.
Recommended settings
| Setting | Value | When |
|---|---|---|
| resolution | 768p (default) | Raise to 1080p or 2k (16:9 or 9:16 only) for a crisper master |
| duration | 5–30 s for text or a first frame; 5–15 s with references | Over 15 s needs prompt_enhancement turbo or quality |
| aspect_ratio | auto (default) | 16:9 for text, the reference's shape with references; leave off with a first frame |
| prompt_enhancement | quality (default) | Balanced; use disabled when your prompt is already long and fully written |
| seed | empty or a fixed value | Fix it to repeat a shot and iterate one clause at a time |
Good to know
- A first frame (the
imageinput) cannot be combined with references, and aspect_ratio does not apply to it — the clip keeps the image's own shape. - References take up to 9 images, 3 videos (only the first 5 seconds of each are read) and 3 audio files, 12 in total; name them
<Picture 1>,<Video 1>and<Audio 1>in the prompt. - 1080p and 2k render in 16:9 or 9:16 only, and clips over 15 seconds need turbo or quality enhancement.
- The model is weakest at long on-screen text, soft organic motion such as petals, paper and hair, and close hand work; let speech carry the steps and keep hands still.
Use it from the API
Send a POST with your prompt and settings, and the response returns the finished clip. This is the default text-to-video example:
# pip install "segmind>=1.1.0"
# export SEGMIND_API_KEY="YOUR_API_KEY"
import segmind
result = segmind.run(
"heygen-video-1",
prompt="A cheesemonger in her fifties, in a navy-and-white striped apron over a grey shirt, stands behind a worn wooden counter stacked with wheels of aged cheese in a small corner shop. Locked-off camera on a tripod, a medium shot framing her from the waist up, no camera movement. Daylight falls through the shopfront window to her left. She lifts one pale wedge of cheese toward the lens, holds it steady, and says to camera: \"This one's aged eighteen months. Taste it.\" Visible skin texture, pores and fine lines on her face and hands; no retouching, no beauty filter, no glamour lighting. Muted colour, no teal-orange grade. Plain apron and shelving, no logos, brand names, printed words or badges anywhere in frame. Sound: her voice close and clear, a short knife tap against the hard rind, a doorbell jingling once in the background, the low murmur of a few customers further back. No music.",
resolution="768p",
duration=8,
aspect_ratio="auto",
prompt_enhancement="quality",
)
print(result.get("output"))# 1. Submit the request; the reply carries its request_id
REQUEST_ID=$(curl -s -X POST "https://api.segmind.com/v2/heygen-video-1" \
-H "x-api-key: YOUR_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"prompt": "A cheesemonger in her fifties, in a navy-and-white striped apron over a grey shirt, stands behind a worn wooden counter stacked with wheels of aged cheese in a small corner shop. Locked-off camera on a tripod, a medium shot framing her from the waist up, no camera movement. Daylight falls through the shopfront window to her left. She lifts one pale wedge of cheese toward the lens, holds it steady, and says to camera: \"This one'\''s aged eighteen months. Taste it.\" Visible skin texture, pores and fine lines on her face and hands; no retouching, no beauty filter, no glamour lighting. Muted colour, no teal-orange grade. Plain apron and shelving, no logos, brand names, printed words or badges anywhere in frame. Sound: her voice close and clear, a short knife tap against the hard rind, a doorbell jingling once in the background, the low murmur of a few customers further back. No music.",
"resolution": "768p",
"duration": 8,
"aspect_ratio": "auto",
"prompt_enhancement": "quality"
}' | jq -r .request_id)
# 2. Check its status until it is COMPLETED
curl -s "https://api.segmind.com/v2/requests/$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"Resources
- Try HeyGen Video on Segmind
- API reference
- HeyGen Video — the model end to end: the three modes, how to prompt, duration and resolution limits, reference labels, and the full parameter and error reference
- Models — the models behind the HeyGen API — how HeyGen Video differs from the Avatar rendering engines and HeyGen Voice
- Create HeyGen Video — API reference — the full request and response schema for the create endpoint, the per-mode rules, reference input formats and validation error codes
- #text-to-video
- #heygen
- #model-launch
- #native-audio
- #reference-to-video
