HeyGen Video: release, price and how to use it

HeyGen launched HeyGen Video, its first general-purpose video model, on 30 September 2026. It is built on MiniMax H3 and post-trained by HeyGen. It makes 5 to 15 second clips with their own sound, from a prompt, a first frame or up to 12 references, at up to 2K. The API lists it at $0.03 a second at 768p. On Versely it has been live since 22 October 2026, from 8 credits.

Key facts

Item	Detail
Released	30 September 2026, by HeyGen
Built on	MiniMax H3, post-trained by HeyGen
Modes	Text to video, image to video, reference to video
Clips	5-15 s, 24 fps MP4 with stereo audio
Resolution	480p, 768p, 1080p, 2K
References	Up to 9 images, 3 videos and 3 audio clips (12 in total)
API price	$0.02/s at 480p, $0.03/s at 768p, 3x that at 1080p and 2K
API model ID	heygen-video-1
On Versely	Since 22 October 2026, 8 to 72 credits a clip

Sources, all read on 23 October 2026: HeyGen's model page, its API changelog and its video model price guide (updated 21 October), which gives the launch date and per-second prices. fal also lists the model as HeyGen Video Text to Video. The quality claims below are HeyGen's, not Versely's own tests.

What it is, and what is new

HeyGen is best known for avatars: Avatar IV and Avatar V animate a presenter, or your digital twin, saying a script. HeyGen Video is a different product. It generates the whole scene from a description (subject, setting, light, camera move and sound) rather than animating a person you already have. HeyGen pitches the two together: your avatar delivers the message, and HeyGen Video supplies the warehouse, the product shot or the city around it.

There is no earlier HeyGen Video to compare it with, so the useful comparison is the model underneath. MiniMax H3 is a 2K model with clips of up to 15 seconds. HeyGen's post-training is aimed at control rather than spectacle, and its model page lists what that means:

• Literal prompts. It "follows a brief literally, so you get dialogue only when you ask for it". Sound effects, room tone and speech come from the same prompt as the picture, with no separate voiceover or lip-sync pass.
• References you can name. In reference mode you address uploads as Picture 1 or Video 1 in the prompt, and HeyGen says the model keeps the colour, form and markings of a product you supply.
• Repeatable seeds. Send the same prompt with the same seed and HeyGen says you get an identical file back, so you can change one clause at a time. Without a seed, every run is a new take.
• Styles. HeyGen's launch films covered twelve styles from one model, from live action and plasticine stop-motion to cel anime and watercolour.

HeyGen has also kept extending it. In October its API changelog added 1080p (billed at the 2K rate), an auto duration that picks 8 to 15 seconds to fit quoted dialogue, and clips of up to 30 seconds for text and image to video. The 30-second limit applies only at the turbo and quality enhancement settings. Reference to video and the ultraquality setting stay at 15 seconds.

Checking the headline claims

"The #1 text-to-video model." HeyGen's model page says HeyGen Video ranks first on Artificial Analysis' text-to-video leaderboard. Arena rankings shift as votes come in, and the board has separate tables for clips with and without audio, so check the live leaderboard for the one you care about.

"The cheapest of the top video models." HeyGen's own price guide puts it at $0.03 a second at 768p with audio, and calls that about 40% below the next model with sound on its list. That list is HeyGen's selection, scored by HeyGen. Note what the rate covers: 1080p and 2K bill at three times the 768p rate, and HeyGen renders those two only in 16:9 or 9:16.

Commercial use. HeyGen says clips can be used for ads, social posts and client work under your HeyGen commercial licence. It also says plain clothing and equipment can pick up invented logos, and suggests adding "no logos, brand names, printed words or badges" to the prompt.

Where it is weaker

HeyGen is unusually frank about the limits. It says the model performs best on "short, contained shots with one subject, one place and one action", and that close hand work is "one of the model's weaker areas". For tutorials, it suggests keeping hands still and letting the narration carry the steps. HeyGen itself says Seedance 2.0 and Veo 3.1 are known for complex, multi-shot cinematic motion, and that HeyGen Video competes on control.

Dialogue has a length budget: HeyGen says written lines fit at about 2.5 words a second, roughly 37 words in a 15-second clip. Anything longer is a job for an avatar or a voiceover.

Who it is for, and who should wait

Good fit:

• Social hooks and transitions in 9:16 that need sound in the clip.
• Product scenes where the product has to look like the one you photographed: use reference mode.
• B-roll and cutaways for founder, explainer and training videos.
• Anyone who tests prompts methodically and wants a seed that really repeats.

Test first, or wait, if:

• Your shot depends on hands, fast multi-subject action or several cuts in one clip.
• You need square or 4:3 output at 1080p or 2K. HeyGen renders those resolutions in 16:9 or 9:16 only.
• You want a talking presenter reading a long script. That is HeyGen's avatar products' job, or a lipsync model's. For the avatar side, see our HeyGen alternatives guide.

How to use it on Versely

Versely runs HeyGen Video in all three modes on one HeyGen Video model page: text to video, image to video, and reference to video with up to 9 reference images. Clips run 5 to 15 seconds at 768p, 1080p or 2K. Text and reference jobs offer six aspect ratios, from 21:9 to 9:16. Jobs start at 8 credits and run up to 72, depending on resolution and length. For comparison, the MiniMax H3 text-to-video row started at 52 credits on Versely on 23 October 2026.

1. Open the HeyGen Video page and choose the mode.
2. Write the shot and the sound together: camera and lens, one subject, one action, and what the scene should sound like. Add "no music" if you want no music bed.
3. Keep dialogue to about 37 words for a 15-second clip, and put it in quotes.
4. For reference to video, call your uploads Picture 1, Picture 2 and so on in the prompt. Our guide to spending H3's nine reference slots applies here, since HeyGen Video is built on H3.
5. Test at 768p, then render the keeper at 1080p or 2K.

Versely runs HeyGen Video at ultraquality, HeyGen's highest prompt-enhancement setting. HeyGen says enhancement can reword your prompt, so the model may not see your exact wording. The Versely page shows no seed control, so each render is a new take. Our guide to seeds in AI video explains what a seed locks and what it doesn't.

You can sign up for free and use it on the web or in the iPhone app with one account.

FAQ

When was HeyGen Video released?

HeyGen launched it on 30 September 2026. It went live on Versely on 22 October 2026.

Is HeyGen Video the same as a HeyGen avatar?

No. Avatars (Avatar IV, Avatar V) animate a presenter or your digital twin saying a script. HeyGen Video generates a whole scene, with its sound, from a prompt, an image or references.

How much does HeyGen Video cost?

On HeyGen's API, $0.02 a second at 480p and $0.03 at 768p with audio, and three times the 768p rate at 1080p and 2K, as read on 23 October 2026. HeyGen's Creator plan, at $29 a month, also includes it. On Versely a clip costs 8 to 72 credits, depending on resolution and length.

Does HeyGen Video make sound?

Yes. Dialogue, room tone and sound effects come from the same prompt as the picture, as stereo audio. HeyGen says you get dialogue only when you write it, at about 2.5 words a second.

What model is HeyGen Video built on?

MiniMax H3, post-trained by HeyGen. The API model ID is heygen-video-1.