HeyGen Video and Voice for Marketers: What They Do and How to Use Them

Key Takeaways

• HeyGen, known for talking avatars, now also sells a general-purpose video model (HeyGen Video) and its own voice model (HeyGen Voice). Both are available through its API.
• HeyGen Video has three modes: text to video, image to video, and reference to video. Every clip comes with its own audio, so there is no separate voiceover step.
• Reference to video is the mode marketers should look at first: you can show the model up to 9 images, 3 videos and 3 audio files so a character, product or place stays consistent from clip to clip.
• Billing is per second of output. Image and audio references are free, a reference video is billed on top, and HeyGen is running a 50% launch discount through the end of October.
• HeyGen Voice comes in three versions. HeyGen's own page lists them at $0 (Turbo Free, until November 30, 2026), $4 (Turbo) and $15 (HeyGen Voice) per million characters.
• In our own tests, the references were copied almost literally, and brand words in a prompt ended up as text on props. Both are easy to avoid once you know. The step-by-step version is in our how to use HeyGen Video guide.

What HeyGen Video is

For years HeyGen meant one thing in marketing: a presenter, real or digital, talking to camera. HeyGen Video is different. HeyGen describes it as its first general-purpose video model, built to make a clip from a text prompt, a still image or reference media, with the model identifier heygen-video-1. According to HeyGen's developer documentation it is built on MiniMax H3 and post-trained by HeyGen.

Three details matter for a marketing team:

• Sound is part of the clip. Each clip carries its own dialogue, ambience and sound effects, generated from the prompt in the same pass. HeyGen's documentation states there is no separate speech or lip-sync step.
• Formats are broad. Resolutions run from 480p to 2K, and aspect ratios include 16:9, 9:16, 1:1, 4:3, 3:4 and 21:9, so one model covers a feed post, a story and a cinema-style banner.
• The file is ready to use. Output is MP4 (H.264) at 24 frames per second with stereo AAC audio.

HeyGen also reports that in blind side-by-side tests (4,800 votes on the Artificial Analysis Arena query set) HeyGen Video ranked first against Seedance 2.5, Seedance 2.0, H3 Max and Kling 3.0. That is HeyGen's own result, so treat it as a reason to test the model, not as proof it will win on your brief.

See it: a clip, its reference images and a voice sample

Here is a five-second clip made on HeyGen Video from only the two reference images shown under it, with no text-to-video step. The prompt asked for a locked camera and one slow push-in. The same face, ears and cape held in both of our test clips.

And here is a seven-second voice sample made with HeyGen Voice from one line of script and one of HeyGen's public voices, Sterling. Press play.

HeyGen Voice sample, voice Sterling: "Your launch is tomorrow, and the video is already done. One script, one voice, ready before your coffee gets cold."

The three modes, in marketing terms

• Text to video turns a scene description into a clip. It is the fastest way to explore ideas, and the least predictable for anything that has to look like your brand.
• Image to video takes your image as the literal first frame and animates it. Use it for product shots, listings and stills you already own.
• Reference to video uses images, videos and audio as guides for the character, motion, style or sound. This is the mode for a recurring character, a mascot, a product that must keep its shape, or a look you want repeated.

Clip length depends on the mode. Text to video and image to video accept 5 to 30 seconds on the standard enhancement settings, and reference to video accepts 5 to 15 seconds, so a 30-second spot is a job for several clips cut together.

What reference to video can take

HeyGen's documentation sets these limits for reference to video:

• Up to 9 reference images, 3 reference videos and 3 reference audio files, with 12 references in total.
• Only the first 5 seconds of each reference video are used.
• References can be a link, an uploaded file or inline data, within size limits (for example 16 MB per image link and 32 MB per video or audio link).

A practical reading: give the model a clean character sheet and one or two shots of the place, not a folder of mood pictures. The more references you add, the more it has to reconcile.

What it costs

HeyGen bills per second of output video. For reference to video, any reference video is billed too (input seconds plus output seconds), while image and audio references are free. Clips at 1080p and 2K cost three times the 768p rate. HeyGen is also running a 50% launch promotion through the end of October, so check its pricing page for the rate that applies when you read this.

In our own tests on a HeyGen Pro account, a five-second reference-to-video clip with only image references cost 5 credits, and one that also included a reference video cost 10. Those are our measured numbers on one plan, not a price list.

HeyGen Voice: three models

HeyGen Voice is HeyGen's own speech model, and it covers public voices and clones through the same endpoints. There are three versions:

• heygen-voice-1 is the standard model. It supports public, instant and professional voices.
• heygen-voice-1-turbo supports public and instant voices and is described as faster.
• heygen-voice-1-turbo-free supports public and instant voices on shared capacity, with no guarantee on speed or availability. HeyGen lists it as free until November 30, 2026.

Cloning works two ways. An instant voice needs one recording (the first 3 minutes are used) and is usually ready within seconds. A professional voice needs 1 to 10 recordings totalling 20 minutes or more, and is trained before it can be used. Speech comes back as a WAV file, and a separate speech-to-speech endpoint can re-speak a recording in an instant voice.

Speech comes back as a WAV file, up to 5,000 characters per request, and HeyGen's default limit is 30 requests a minute per workspace member.

What it costs. HeyGen's developer documentation publishes no rates, but HeyGen's own Voice Turbo page (last updated October 21, 2026) lists them per million characters: $0 for Turbo Free until November 30, $4 for Turbo and $15 for the full HeyGen Voice. A short ad script of 500 characters is therefore about a fifth of a cent on Turbo and under one cent on the full model. HeyGen says it will give advance notice before the free offer changes, and it also reports that 64% of its users preferred Turbo over ElevenLabs in blind listening tests. That is HeyGen's own claim.

For marketers the point is simple: one approved brand voice, reused across ads, product videos and support clips, instead of re-recording every time the script changes.

What we learned testing it

We ran reference-to-video clips on a HeyGen Pro account on October 27, 2026. Four lessons are worth passing on:

1. References are copied almost literally. We gave it a rough 3D mannequin as a stand-in for a person, and the mannequin came back in the clip. Use a finished character image, and keep place references free of any figure.
2. Do not name a camera brand in the prompt. Writing a professional camera maker's name made that name appear as a logo on a sign and a cereal box. Saying "shot on a 35mm cinema camera" gave the same look without the text.
3. Say "blank" for props. Asking for "no readable brand names" still produced gibberish lettering on packaging. Asking for plain, unmarked boxes worked better, and it is still worth checking every frame.
4. Image-only clips are quick. Several five-second clips ran in parallel and each finished in roughly one to three minutes.

A simple workflow for a marketing team

1. Write the idea in one sentence, with a character or product at the centre.
2. Make a character sheet first: one full-body view and one close-up, on a plain background.
3. Run two or three short reference-to-video clips with those references and one place each.
4. Add one approved voice line with HeyGen Voice if the clip needs narration.
5. Cut the clips together, then check every frame for stray text before it goes out.

Where MITPO fits

MITPO's Creative Studio already lets you pick a video model and attach reference images, and HeyGen Video and HeyGen Voice are being added to its video and speech pickers. Want the hands-on version first? Read how to use HeyGen Video, with the prompt formula we use. The aim is the same workflow without leaving one place: references in, clips out, and a preview of what each model does before you spend a credit.

FAQ

Is HeyGen Video the same as HeyGen's avatar videos?

No. Avatar videos are built around a presenter speaking to camera. HeyGen Video is a general-purpose model that makes scenes, products and characters from a prompt, an image or reference media.

Does HeyGen Video make its own sound?

Yes. HeyGen states that each clip includes dialogue, ambience and sound effects generated with the picture, with no separate lip-sync pass.

How long can a clip be?

Reference to video accepts 5 to 15 seconds. Text to video and image to video accept 5 to 30 seconds on the standard enhancement settings.

Can I clone my brand voice?

HeyGen Voice supports instant clones from one recording and professional clones from several recordings of the same speaker. Only record voices you have the right to use, with the speaker's consent.

Is HeyGen Voice Turbo free?

HeyGen lists the Turbo Free model at $0 per million characters until November 30, 2026, on shared capacity with no guaranteed speed. The standard Turbo model is $4 per million characters and the full HeyGen Voice model is $15, according to HeyGen's own page.

Which HeyGen voice model should I use?

Use Turbo for quick drafts and everyday voiceovers with public or instant voices. Use the full HeyGen Voice model when you need a professional voice clone, because Turbo cannot speak those.

Is the free voice model safe to build a campaign on?

The turbo-free model is on shared capacity with no guaranteed speed or availability, and HeyGen lists it as free until November 30, 2026. Use it to test, and plan for the paid models in production.