5 prompts · AI avatar video
Prompts and scripts for AI avatar and talking-head video
You are generating a presenter-led video with a synthetic avatar and want it to read as competent rather than robotic.
Avatar tools render whatever script you give them, which means the script carries the entire performance. Sentence length controls pacing, punctuation controls breath, and explicit gesture cues control the dead-eyed stillness that gives avatars away. These prompts write for that constraint.
Training, onboarding, and internal comms videoMultilingual product explainersCreators who want a presenter without filming
The prompts
1. Avatar-ready presenter script
Claude or ChatGPT, then Synthesia or HeyGenA script written for synthetic delivery
Write a [MINUTES]-minute presenter script about [TOPIC] for [AUDIENCE], to be delivered by an AI avatar.
Constraints:
- Maximum 18 words per sentence. Vary length deliberately — a short sentence after two long ones creates rhythm.
- One idea per paragraph, each paragraph 2-4 sentences, so the avatar has natural reset points.
- No tongue-twisters, no clustered sibilance, no words with ambiguous pronunciation.
- Mark [GESTURE: ...] and [PAUSE] cues between paragraphs.
- Conversational register: contractions, direct address, no corporate abstractions.
In practice: The 18-word cap is the single biggest quality lever in avatar video.
2. Localisation pass
Claude or ChatGPT, then Synthesia or ColossyanOne script, many languages, without literal translation
Localise this avatar script into [LANGUAGES]. Do not translate literally.
For each language: adapt idioms and examples to something locally familiar, adjust formality to the local business norm, keep sentences under 18 words after translation, and flag any term that should stay in English (product names, technical terms).
Note where the runtime will change significantly — some languages run 20-30% longer.
Script: [PASTE SCRIPT]
In practice: The runtime flag matters — a 3-minute English script can become a 4-minute German one.
3. Scene and slide plan
Claude or ChatGPTDeciding what sits beside the avatar on screen
Break this avatar script into scenes. For each: the spoken lines, the avatar's framing (full body, waist up, close), the on-screen support (slide, screen recording, chart, b-roll, or nothing), and the 3-6 words of on-screen text.
Rule: never leave the avatar alone on screen for more than 20 seconds without a visual change.
Script: [PASTE SCRIPT]
In practice: The 20-second rule is what stops avatar video feeling like a hostage tape.
4. Objection-handling module
Claude or ChatGPTSales and product avatar video
For a product avatar video about [PRODUCT] aimed at [AUDIENCE], write a 45-second segment that names the strongest objection a sceptical buyer has, states it plainly in their words, and answers it with one concrete proof point.
Do not soften the objection. Do not answer with a feature list. Do not claim anything I have not given you.
Proof points available: [LIST]
In practice: Naming the objection out loud outperforms avoiding it, especially in a synthetic delivery.
5. Pronunciation and pacing pass
Claude or ChatGPTFinal cleanup before rendering
Prepare this script for avatar rendering. Output a table of every proper noun, acronym, number, and unusual word with a phonetic spelling. Then return the full script with pauses marked, emphasis words bolded, and any sentence over 18 words split.
Do not change meaning or add claims.
Script: [PASTE SCRIPT]
In practice: Run this last, right before you paste into the avatar tool.
Tools that finish the job
Where these prompts go next
A prompt produces the script, the shot list or the markup. These are the tools that turn that output into a finished video.
Synthesia
Studio-quality AI presenters for training and internal comms.
Starts at $22/mo
View tool profile →HeyGen
Humanlike avatars and talking head ads without a studio.
Starts at $29/mo
View tool profile →Colossyan
AI video platform built for workplace learning and L&D teams.
Starts at $28/mo
View tool profile →