Roughly is the operative word. Localisation done carelessly looks cheap in every market at once. Here is how to do it properly without burning credits.

Why is a straight machine translation not enough?

Because speech is not just words. Languages expand and contract — German runs long, some languages compress. If the translated script is 20% longer than the original, the avatar must either speak faster or the video must change length. Neither is free. Idioms also break, and a joke that lands in one market can read as a literal, baffling sentence in another.

Localisation is translation plus timing plus tone adaptation. Skip any of the three and the result feels foreign.

The localisation workflow

1. Lock the source script first. Re-translating because you edited English is wasted credits.
2. Export the transcript and translate it with a translator who knows the product, not raw machine output.
3. Compare lengths per paragraph and rewrite the ones that blow up.
4. Pick a voice native to each market instead of reusing the English voice accented.
5. Render one language end to end and check lip sync and timing before doing the rest.
6. Batch the remaining languages only after that single render passes.

Step five is the discipline that saves money. Rendering twenty languages at once means twenty refunds if the timing is off.

How do the credits add up?

You pay per minute of generated video, per language. A ten-minute training module in ten languages is one hundred minutes of avatar render at Avatar IV rates, which is a real budget line. Trim the source: cutting a bloated English video by 30% before localising cuts every downstream language by the same 30%.

What breaks first in translation?

Numbers, dates, and currency formats. Rendered literally, they are wrong or confusing per market. Acronyms spelled out in English get read as nonsense in another language. And humor nearly always needs rewriting rather than translating. Have a native speaker watch each finished version — not to check the words, but to flag the moments that feel off.

Who should invest in this pipeline?

Teams selling into multiple markets with a repeatable onboarding or training video, where one good localisation process pays off across dozens of assets. It is overkill for a single one-off video in one market — the setup cost, both in credits and editing, will outweigh the benefit.

Key Takeaways

• Localisation is translation plus timing plus tone, not a straight text swap.
• Lock your source script before translating; edits later cost double.
• Credits bill per minute per language, so trim the source first.
• Numbers, dates, acronyms, and humour are the first things to break.
• Render one language fully as a test before batching the rest.

The pitch is obvious: no camera, no studio, no second take. The reality is good enough for training videos and dull social content, and not good enough to pass as a live recording.

What is HeyGen Avatar IV?

It is the fourth generation of HeyGen’s avatar engine. Compared with earlier versions it adds more natural upper-body motion and better lip sync on fast speech. It works from either a stock avatar or your own uploaded photo, and supports many languages with voice options including cloned voices.

The output is a video file, usually rendered at 1080p, with the avatar speaking over a background you choose.

How do you make an avatar video step by step?

1. Create a HeyGen account and open Avatar IV.
2. Pick a stock avatar or upload a clear frontal photo of yourself on a plain background.
3. Type or paste the script, or generate a draft from a URL.
4. Choose a voice and language, and test a short line first.
5. Pick a background — solid color, upload, or a HeyGen template.
6. Render. A one-minute clip takes a few minutes.

Test with one sentence before rendering the full script. If the lip sync or voice is wrong, you have saved yourself a wasted render.

What does it cost?

HeyGen runs on a credit system. A free tier offers a small monthly allowance, typically enough for a couple of minutes, and paid plans scale credits by the minute. Avatar IV generation costs more credits than the basic avatars, so a five-minute training video can eat a meaningful chunk of a mid-tier plan.

Check per-minute credit costs before scripting a long video, not after.

What does the avatar still get wrong?

Hands. They appear only in short, generic gestures and never hold or manipulate objects convincingly. Emotion is narrow — the face reads friendly and neutral well, genuine surprise or anger poorly. Long pauses and unusual phrasing, like spelled-out acronyms, can break the illusion. And the eyes do not track a camera precisely, so a close-up shot looks slightly off.

Who should use it?

Companies making onboarding and product training videos where a consistent presenter matters and a reshoot is expensive. Localising content into 20 languages without rerecording. Solo creators who dislike being on camera. Skip it for anything needing genuine performance — comedy, emotive storytelling, or hand-heavy demonstrations.

Key Takeaways

• HeyGen Avatar IV animates a single photo or stock avatar into a talking video.
• It supports many languages and cloned voices for translation without reshoots.
• Credits bill per minute; Avatar IV costs more than basic avatars.
• Hands, strong emotion, and close-ups of the eyes are the weak spots.
• Good for training and localisation; not for performance-driven content.