How to Write the Script for an AI Influencer Video
A 10 second video is about 25 spoken words, so the script is the whole video. Here is how to write one that sounds spoken, how to have it generated from a direction, and what to fix by hand.
A short-form talking-to-camera video is mostly its script. There is no b-roll to hide behind and no edit to save a flat opening, so the words are the product. In HexUGC the script is also the instruction the rest of the pipeline follows: the voice reads it, the lip-sync is timed to that audio, and the captions are cut from the same audio. Get the script right and the video is largely decided.
Twenty-five words is a whole video
Spoken delivery runs at roughly 2.5 words a second. That makes a 10 second video about 25 words, a 20 second one about 50, and a 30 second one about 75. Most first drafts written for the page come in at double that and then get read too fast to follow.
Write to the count on purpose. One idea, one turn, one line that lands. If you have more to say than fits, that is a second scene rather than a longer paragraph, and scenes stitch into one file. We covered that in making an AI video longer than 10 seconds.
Three ways to get the words
You are not obliged to start from a blank box. The create wizard takes the script three ways:
- Paste one you already have. If you write your own hooks, nothing here gets in the way.
- Write it in the box. Type directly and move on.
- Give a direction and have it drafted. Describe the video in plain language, who you are talking to, what it is about, the angle you want, and a draft comes back written to length. It lands in an editable field, not in the render.
The draft is a starting point with a deadline, not an oracle. It is reliably good at structure and length discipline, and reliably generic about your specifics. Read it once, cut the first sentence if it is throat-clearing, and put your own detail back in.
The first line is the job
Whatever you do with the rest, spend your attention on the opening. The feed decides in about a second, and a generated draft will happily open with a warm-up clause that costs you that second.
Name the person you are talking to, or state the outcome, or start mid-thought. Avoid "in this video I am going to", which is a phrase nobody has ever said out loud. We broke down what separates a clip that holds from one that scrolls past in the anatomy of an AI video that gets 20k views.
Write it to be spoken, not read
Everything you type gets spoken, so a script is not a document.
- No stage directions. Brackets, "(laughs)", emoji and asterisks are text to a voice model, not instructions.
- Spell out anything you would say differently from how it is written. Write "twenty eight" if that is what should be heard, and write out an abbreviation you would say in full.
- Punctuation is your pacing. Full stops make pauses, commas make small ones, and a long unpunctuated sentence comes back as a rush.
- Read your draft aloud before you generate. If you stumble, the voice will too. More on choosing and keeping the voice in picking a voice for your AI influencer.
The script is also the captions and the price
Captions are cut from the rendered audio and burned into the video word by word, so a script that sounds right captions right. Nothing is typed twice.
It is also the meter. Length is estimated from the script, and the clip is billed by the second, so a 10 second voiced video costs 28 credits. Trimming a rambling draft is not only an editing decision.
What this does not do
The script step drafts, it does not strategise. It does not know your offer, your audience or what has worked on your account, and it will not tell you that your angle is weak. That judgement stays yours.
It also writes one script at a time. Generating ten variants of a hook to test against each other sits on our roadmap rather than shipped, so variant testing here means running the videos one after another and deciding yourself.
And there is no free tier to draft against. You pay from the first generation, with Starter at $15 a month working out at roughly 10 videos and Creator at $30 at roughly 21.
Closing
Write twenty-five words you would actually say, then let the character say them.