How to Make AI Avatar Videos: Scripts, Voiceover and Consent

What an AI avatar video is, and when it is a good fit

An AI avatar is a digital presenter, realistic or cartoon-style, that appears on screen and reads your text in a synthetic voice. Tools built for this include HeyGen and Synthesia, and some general video generators can do parts of it. What each offers, with which limits and terms, changes often, so check the current details before you pick one.

It fits when you need a consistent face and voice and can't or don't want to film yourself: explainers, internal training, product walkthroughs or FAQ videos. It fits less when the value of the video is the real person, such as a personal testimonial, an opinion or a sensitive message.

• Good fit: repeatable explainers, onboarding videos, multi-scene walkthroughs, updates you will re-record often.
• Poor fit: testimonials, apologies, announcements where people expect a real human, anything that depends on trust in a specific person.

Your script does the heavy lifting

An avatar doesn't improvise or correct itself: it says exactly what you wrote. That makes the script the part you control, and it has to be written for listening, not reading.

• Short, direct sentences; long paragraphs make an avatar sound robotic.
• Numbers, acronyms and names written the way they are pronounced, or with a pronunciation note.
• Marked pauses where the topic changes, so the video can breathe.
• One idea per scene, with on-screen text kept separate from what the avatar says.
• No claims you can't back up: an avatar adds an air of authority even when the content hasn't earned it.

Act as a scriptwriter for business videos.
Write a script for a [1-2 minute] video in which a virtual presenter explains [topic] to [audience].
Tone: [friendly and professional].
Format: a table with scene, what the presenter says (written as it is pronounced) and what is shown on screen (text, image or chart).
Include a short introduction, 3 content blocks and a close that tells viewers [what to do next].
Limits: sentences under 20 words and no invented facts. If information is missing, leave a [bracketed placeholder] instead of filling it in.

Voice: getting a natural voiceover

The voice is what most gives away an AI-made video, and also what you can improve the most. A good voiceover starts with the text: ask the AI to prepare it with pause and emphasis marks, then listen to how it sounds in your voice tool.

Prepare this text to be read by a synthetic voice in [US / UK] English.
Write numbers and acronyms the way they are said, split long sentences, add [short pause] and [long pause] where useful, and put the words that need emphasis in CAPITALS (no more than one per sentence).
Don't change the meaning or add information.
At the end, list any words likely to be mispronounced and suggest a phonetic spelling for each.

[paste your text here]

Listen to the full audio before you generate the video. Fixing a word in the text takes seconds; redoing a finished video does not.

Consent and disclosure: what not to skip

The delicate part of avatars isn't the technology; it's whom they represent and how they are presented. This is not legal advice. Check the tool's terms and the official sources that apply to you.

• Don't create an avatar or voice that imitates a real person without their clear permission. That includes celebrities, colleagues and your former boss.
• If you use your own likeness or voice, or a teammate's, make sure there is consent and that everyone understands what it will be used for and for how long.
• Read the tool's terms on who owns the avatar, how your data is stored and which uses are banned.
• Say the presenter is virtual, or the video AI-generated, wherever the platform or the law requires it. Even where it doesn't, transparency is good practice.
• Never present an avatar as a real customer, a credentialed expert or a genuine testimonial when it isn't one.

Mistakes to avoid

• Reusing text written to be read: it sounds stiff in an avatar's mouth.
• Skipping the pronunciation check on names, brands and figures.
• A long video with no scene changes or visual support: it tires viewers quickly.
• Not saying the presenter is virtual when it would matter, so the audience feels misled on finding out.
• Using a face or voice without the rights or permission to do so.

PromptNow templates for avatar scripts and voiceover

PromptNow's AI avatar spokesperson script template asks for your topic, audience, length and tone in a form and generates a prompt that produces a scene-by-scene script. It is the starting point before you open the avatar tool.

The AI voiceover script template builds the prompt for a text designed to be spoken: pauses, rhythm and pronunciation. It doesn't create the video or the voice; you do that with the tool you choose, and you review and approve the result.