HeyGen Script Tips: Pauses, Pronunciation and Pacing

An avatar reads exactly what you type, which is both the strength and the weakness of avatar video. A script written for the eye, long sentences, symbols, abbreviations, stumbles when it is spoken by text-to-speech. HeyGen gives you a handful of tools to control rhythm and pronunciation, and its help centre is specific about how to use them. This guide puts them in one place, with the writing habits that make the tools work.

Structure the script in scenes

In AI Studio each scene has its own script, avatar, voice and background. Click Add Scene, choose Blank or From Template, and type into the script editor. Use scenes the way an editor uses cuts:

• One idea per scene. It keeps each piece easy to re-record and lets you change the layout with the topic.
• Short scenes for emphasis. A one-line scene lands harder than the same line buried in a paragraph.
• Silent beats. HeyGen suggests adding a scene containing only a pause to create a silent, still avatar between sections.

You can also download and upload a whole script from the top-left menu, pauses included. When uploading, each line, paragraph or spreadsheet cell becomes a scene, segments over 1,000 characters are split automatically, empty lines are skipped, and .doc files are not supported (save as .docx). HeyGen notes that the file is read in your browser, not uploaded to a server.

Pauses and punctuation

Use the Pause button rather than stacking punctuation ("..." or ",,"), which voices interpret inconsistently. Read the script aloud yourself first; wherever you naturally breathe is where a pause belongs.

Pronunciation and the Brand Glossary

Preview the script. When a word comes out wrong, double-click it and select Pronunciation, then type how it should sound, using hyphens to split syllables. HeyGen's examples: "AI" as a-eye, "AWS" as a-double you-s. Useful for brand names, people's names, acronyms and jargon.

Two details to know. HeyGen's Script Tips article says pronunciation rules you apply are saved to your Brand Glossary for future videos. Its AI Studio guide adds that a correction made in the editor applies to that instance in that scene only, and recommends the Brand Glossary itself for consistent pronunciation across a whole project. The safe habit: add every recurring name to the Brand Glossary directly.

Numbers, dates and symbols

Text-to-speech engines stumble on numeric formats. Write them the way they should be said. HeyGen's own examples:

• "2012" as twenty twelve.
• "3/8" as three eighths of an inch.
• "01:18" as one minute and eighteen seconds.
• "10-19-2016" as October nineteenth, two thousand sixteen.

Do the same for symbols, units and web addresses: "per cent", "kilograms", "dot com". British audiences may expect "the nineteenth of October"; write the form your audience says.

Language and spelling

Spell every word correctly, because the voice reads what is there. Do not mix languages within one script: HeyGen warns that inserting Arabic or Chinese words into an English script can cause unpredictable output, and recommends separate scenes or projects per language. To make another language version, HeyGen's Text Editor can translate a script, and video translation handles finished videos; see translation and lip sync.

Length and pacing

Spoken scripts run slower than you expect. As a rough guide, conversational narration lands somewhere around 130 to 160 words a minute, so a one-minute explainer is shorter on the page than it feels. Time a read-through aloud before you build scenes, and cut rather than speed up the voice: a crowded script with a fast voice sounds rushed in any language.

Writing for the ear

1. Short sentences. One clause, sometimes two.
2. Active voice and plain words. "Click Save" beats "the Save option should be selected".
3. Signpost. "First", "next", "finally" help listeners who cannot scan back.
4. Repeat the key point at the end; listeners need it more than readers.
5. Preview every scene. Fix rhythm with pauses and words with Pronunciation before you render.

When text alone will not carry the tone, use Direct Voice to set a delivery such as Calm or Excited, or Voice Mirror to record the line yourself; both are covered in how to clone and direct your voice.

A review checklist before rendering

• Every scene previewed at least once, start to finish.
• All names, products and acronyms checked, and the recurring ones added to the Brand Glossary.
• Numbers, dates and units written out as spoken.
• No mixed languages within a scene.
• Pauses placed where you would breathe, not where the punctuation happens to fall.
• The script saved or downloaded with the project, so a later correction starts from the approved text.

Rendering costs credits, and a wrong name in a finished video means rendering again. Five minutes of previewing is the cheapest part of the job.

Try it

Welcome to the spring update. [pause]
First: the new booking page. It now shows every free slot for the next thirty days.
Next: payments. You can pay by card or bank transfer, and receipts arrive straight away.
That's it for this update. Thanks for watching.

Paste each line as its own scene and replace [pause] with the Pause button set to one second. Preview, fix any name with Pronunciation, add it to the Brand Glossary, then compare with the same text pasted as one long scene.