How to make AI sales clips: steps, tools and rules (2026)
Making a product-selling clip with AI takes five steps: write the sales script, generate the voice, generate the presenter, add products and subtitles, then post it or take it live, using four tool types in a chain. A typical shop makes its first clip in a day. A clip with a realistic talking person needs consent from the owner of the face and voice, and every clip must be labelled as AI. This guide covers each step, how to choose tools, what AI still cannot do, and shows a real clip of AI Watt, made from a real member of our team.
Three kinds of AI sales clip: pick the kind before the tool
Product footage with an AI voice-over
Your own product photos or video, an AI voice reading the script, subtitles on top. The easiest kind, needing only a voice tool and an editor. Fits short clips promoting one product at a time.
A stock character talking
A ready-made character from an AI avatar tool speaks the script in a synthetic voice. Professional-looking but not your shop's person. Fits explainers and announcements, not brand familiarity.
An AI host made from your shop's real person
An AI made from the face and voice of a consenting owner or employee, speaking the sales script in that person's cloned voice. Customers see the same person in every clip and every live. This is the kind that sells steadily, and the kind UpperLive uses; see the example at the end.
The five steps
Write the script from product facts
Start with a table of products, prices, promos, selling points and common questions, then let AI draft a 30–60 second script per product. A structure that always works: open with the customer's problem, introduce the product, give two or three selling points, state the price and promo, close with a call to order. Never let AI invent claims; every sentence about results must come from the shop, and supplements, cosmetics and other regulated products need their wording checked against the rules first. Read the script aloud once: where you stumble, the listener will too.
Generate the voice
Two routes: a ready-made synthetic Thai voice, fast and permission-free, or a clone from a few minutes of a real person's speech, which gives a voice that is truly the shop's. Listen for how numbers and prices are read in Thai, pauses, and loanwords that are often misread. Spell numbers out as words in the script if the tool misreads them, and pick a slightly slower pace than feels natural, because viewers watch the product while they listen.
Generate the presenter
For the third kind, record a source video of the real person as the tool specifies: usually a few minutes, face on, well lit, speaking naturally. The tool then builds an AI host that can speak any script in that person's face and voice. Every trustworthy tool includes a consent step here, where the face owner says a confirmation line or verifies identity. Complete it and keep the record. Do not skip it by using a tool that does not ask; that is the tool that can build from anyone's face, and platforms will treat the result as a deepfake.
Add products, prices, subtitles and the AI label
Edit the presenter clip: cut to product shots as the script mentions them, put prices and promos in large text because many people watch muted, add Thai subtitles throughout, and place an AI-generated label in the picture from the first second, not only in the post caption. Use vertical 9:16 for TikTok, Reels, Shorts and clips inside shopping apps.
Post it, or take it live
The clip can be posted as a short video at once; also tick the platform's AI-generated content option. Using it for live selling needs a system that sequences dozens of clips into a programme, feeds the picture into the shop's live channel for hours, and answers comments during the live, which is beyond ordinary clip tools. How the live system works is explained in What is AI live commerce.
Four tool types and how to choose
We avoid naming brands on purpose, because tools in this category change monthly. Choose by the features in the table instead.
| Type | What it does | Must have | Watch out |
|---|---|---|---|
| Script | Drafts sales scripts from product data, makes variants | Accepts product tables, adjustable tone | Invented claims: a person checks every script |
| Voice | Reads the script in Thai, or clones a real voice | Clear Thai, correct numbers, a consent step for cloning | Cloning someone who has not consented breaks tool rules and the law |
| Presenter (AI avatar) | Builds an AI host from real video, or animates a stock character | Lip sync with Thai, natural gestures, high-resolution vertical export, face-owner verification | Tools that never ask for consent, and lips that do not match Thai words |
| Editor | Adds products, prices, subtitles, the AI label, resizes | Automatic Thai subtitles, vertical templates, fast export | Thai fonts with floating vowels, and copyrighted music |
Most shops doing it themselves rely on the voice and editing tools, and outsource only the AI host when they want the brand's own face and voice, because that step needs a good source recording, several rounds of settings and a consent process.
Three rules that keep AI clips selling without trouble
- Consent — the face and voice come from a person who consented in writing, stating which brand, for how long and how to withdraw. Record it even for the owner. Never use a celebrity or anyone unaware.
- Labelled as AI — every clip and every live tells viewers clearly that it is AI-made, in the platform's option and on screen. TikTok, Facebook and YouTube have rules on this, and the rules change.
- Only true statements — prices, promos and claims match what the shop confirmed; regulated products get their wording checked; a human team controls every script before air and can stop or change it the moment information changes.
What AI still cannot do in a sales clip
- Handle or demonstrate the product — an AI host talks and gestures but holds nothing; products are shown through photos and footage.
- Answer new questions on its own — a clip is a prepared script; anything outside the answer set needs a person to type it.
- Guarantee sales — a good clip helps people understand the product faster; sales depend on product, price and reach.
- Replace the owner — customers who buy because they trust the owner still want the real one at times; AI covers the hours the owner cannot.
A real example: AI Watt, made from a real team member
Watt is a real member of the HelloAds team who consented to an AI version of his face and voice. We recorded a source video in the studio, cloned the voice and built an AI host that speaks clear Thai and English. The clip below is the result: fully AI-generated from a script the team wrote, labelled as AI under the rules, and the same kind of clip that sells live for shops every day.
Do it yourself, or have a team do it
Doing it yourself pays when you need a few short clips a week and someone in the shop enjoys tools; the cost is several subscriptions, learning time and checking every clip. A team pays when you need a steady volume, a well-tuned brand AI host, or daily live selling, which needs a clip sequencer, a comment answer set and a human team in control backstage for the whole live, not just clips.
If what you want is daily live selling without sitting there yourself, see what the service covers at Hire an AI live host, compare it with hiring a person in Hiring a live host vs an AI host, and if you are setting up your own stream, read Live selling equipment alongside.
In short: an AI sales clip starts with a correct script, a voice that reads prices clearly, a presenter who gave consent, a visible AI label, and a person checking before every release. Tools change; these four are what make a clip sell without trouble later.
Questions this article raises
What software do I need to make AI sales clips?
Four tool types in a chain: a script tool (AI drafts, a person edits), a voice tool (a cloned or synthetic Thai voice), a presenter tool (an AI avatar from a real person, or a stock character) and an editor for products, prices and subtitles. Most shops start with voice and editing; a full talking-presenter clip needs an AI avatar tool with a consent step for the face owner.
Can I use my own or an employee's face, and do I need permission?
Yes, with written consent from the owner of the face and voice every time, even if that is you, because most AI avatar tools require identity verification before creating one. Never use a celebrity, an actor or anyone unaware; it breaks the tool's rules, the platform's rules and the law.
Must AI clips be labelled as AI?
Yes. The main platforms in Thailand, TikTok, Facebook and YouTube, require AI-generated or AI-altered content to be disclosed, especially realistic talking people. Label it both in the platform's own option and on screen. Skipping it risks reduced reach or removal, and lost trust when customers find out later.
Can an AI clip be used for live selling, or only for posts?
Both. Pre-made clips can be posted as short videos at once. Going live needs a system that sequences clips into a programme, sends the picture into the shop's channel, and answers comments, which is what an AI live commerce service provides, with a human team in control of the script and comments backstage.
Doing it yourself versus hiring it out: what is the difference?
Doing it yourself suits a few short clips a week and a shop with someone who enjoys tools; the cost is several subscriptions and learning time. Hiring suits a steady volume of clips, a brand-specific AI host, or daily live selling, which needs a system and a backstage team rather than just clips.