HeyGen for ads: a practical guide

What HeyGen does for ad work

HeyGen makes two kinds of video: avatar videos, where a stock avatar, custom avatar or photo reads your script with lip sync, and HeyGen Video, a scene model for short clips with their own audio. For ads, the avatar side covers talking-head reads and the scene side covers inserts.

The avatar engines are chosen with a setting: "Choose Avatar III, IV or V with engine (Avatar IV by default)" (HeyGen API). HeyGen Video is "built on MiniMax H3 and post-trained by HeyGen" (HeyGen Video docs).

Two specs matter most for direct-response work. The 4:5 avatar ratio lets you render a feed version without cropping, and translation lets one tested script run in other languages with the same presenter.

Which ad formats HeyGen suits

HeyGen suits formats built on one person speaking to camera. The table is guidance from published capabilities, not test results.

HeyGen is a weaker fit for drama, animation and brand film. Those need longer scene takes, several characters moving in one shot, or a stylised look, which the 15-second scene model limit and avatar format do not cover. Use a video model such as Kling or Runway for those.

Workflow: from script to finished HeyGen ad

Write for the ear, render the read, then build the ad around it in an editor. A 30 to 45 second avatar ad is usually one avatar read plus three to six inserts.

1. Lock the script. Time every line aloud. Mark where cutaways go, because you will cover avatar jump cuts with B-roll at those points.
2. Choose the presenter. Use a stock avatar under HeyGen's terms, or a custom avatar of a person who has given written consent. Pick a look that matches the buyer, not the brand team.
3. Set the voice. Pick one voice, or upload a recorded voice track for the avatar to lip sync to. Keep the same voice across every version of the script so tests compare hooks, not voices.
4. Render the read in the final ratio. Render 9:16 for Reels, Stories and TikTok, and 4:5 for the feed. Rendering each ratio avoids cropping off hands or the product.
5. Make the inserts. Use HeyGen Video image-to-video for product shots, or bring in clips from another model.
6. Review every take. Check mouth shapes on "p", "b" and "m", teeth and tongue, hand position, blinking rate, and whether the voice stresses the product name correctly.
7. Edit. Cut out pauses, add captions, insert B-roll over every cut, then add the end card. Put prices, offers and claims in as editor text.
8. Version. Swap one variable per version: hook line, avatar, setting or language.

The hook is usually the only part you need to re-render. Keep the body read fixed and render five to ten hook lines as separate short clips, then join each to the same body.

Translating a tested ad into other languages

Translation is the HeyGen feature with the least competition in ad work: paid plans list "175 languages and dialects" and video translation and dubbing (HeyGen pricing). Use it only on a script that has already been tested in its first language.

1. Translate the script first, in text. Have a native speaker check the hook and the offer. Literal translations of slang hooks often read wrong.
2. Check length. Some languages run longer than English. If the translated read runs over, cut words rather than speeding up the voice.
3. Replace on-screen text. Captions, prices and the end card are editor text, so rebuild them in the new language and currency.
4. Re-check claims locally. A claim approved in one market may not be approved in another. Ask the brand before you ship.
5. Review lip sync on the translated take. Watch it with the sound on and again muted.

Script structure for avatar ads

Avatar reads work best with short sentences, written pauses and no stage directions in the spoken text. The templates below are written for this guide. Square-bracket items are placeholders.

UGC avatar read (30 to 40 seconds, 9:16)

HOOK (0-3s): [one-line problem the buyer says out loud]
PAUSE.
SETUP (3-10s): I tried [old solution]. [Why it failed, one sentence].
TURN (10-18s): Then I found [product]. It [one specific thing it does].
PROOF (18-28s): [Documented claim or approved review, word for word].
CTA (28-35s): [Offer in plain words]. Link's below.

Expert explainer read (45 to 60 seconds, 4:5)

HOOK: Most people [common mistake with the problem].
POINT 1: Here's why that doesn't work. [Mechanism in one sentence].
POINT 2: What does work is [approach]. [One supporting fact the brand can source].
POINT 3: That's what [product] does. [Feature] means [plain benefit].
CTA: If [qualifier], try [product]. [Offer].

HeyGen Video product insert (image-to-video, 5 seconds)

First frame: [product image] on [surface], label facing camera.
The camera pushes in slowly. [One physical detail: cap turns / pump presses / steam rises].
Soft daylight from the left, shallow depth of field.
Audio: room tone and one [sound], no dialogue.

HeyGen Video podcast wide shot (reference-to-video, 8 seconds)

References: [host A image], [host B image], [studio image].
Two hosts at microphones in the studio, medium wide shot.
Host A laughs and leans back; host B nods and points at the table.
No speech. Audio: low room tone, a chair creak.

Use the podcast wide shot as a cutaway between two avatar reads. It hides the switch between hosts and makes the set feel like one room.

Limits, terms and policy notes

The main limit for ad work is the plan: free HeyGen output cannot be used in ads. Check the plan before you deliver anything to a brand.

Commercial use. For Creator, Pro and Business users, HeyGen's terms say "you own all rights in your User Input or User Output" and HeyGen "does not restrict your ability to use User Output for your own purposes (including for commercial purposes)" (HeyGen terms). Free-plan output is licensed "solely for personal, non-commercial, and internal evaluation purposes" and "may not be ... used in connection with commercial activities, advertising, client work". The terms also require you to disclose AI use where the law requires it.

Pricing, as published (USD).

Source: HeyGen pricing. We did not find a credit cost per avatar video on that page when we checked, so test a few renders before quoting a brand.

Platform labels. TikTok counts "Making the primary subject say something they didn't actually say, using AI voice-cloning" as significant AI modification and requires the AIGC label or your own clear disclaimer (TikTok ads policy). Meta labels ad images "created or edited using third-party AI tools with AI info" (Meta). Full detail: AI-generated content labels on Meta and TikTok ads.

Faces and voices. TikTok does not allow AI content using "the likeness of adult private figures used without their permission" (TikTok Help Center). Get written consent before you build a custom avatar or voice clone from anyone, and never use a public figure.

Testimonial wording. An avatar is not a customer. Do not script it to claim personal results nobody reported, and do not present it as a doctor or other expert unless the brand has an approved claim and a real person behind it.

HeyGen or a skilled AI ad editor

Every page that ranks for avatar ads is written by an avatar vendor, so none of them answers when to use the tool yourself and when to hire someone. The answer depends on the script and the formats you need.

Many working editors use both: HeyGen for the talking reads, and a video model for the scenes around them.

HeyGen in a one-script, every-format workflow

HeyGen fits the one-script method as the talking-head and translation step. Keep the script, the voice and the hook list fixed, render the avatar version, then reuse the same voice track for a podcast cut or an animated version built in another tool. The method is in one winning script, made in every ad format.

Anglr pays AI creators, editors and agencies on performance to remake winning ad scripts in every format: UGC, podcast, drama, animation and brand film. Anglr provides the winning scripts and angles, and every video gets a QC and compliance review before it goes live. If you already make full ads with HeyGen and other tools and know they perform, see how it works for creators.