How to make a product demo video with AI

A good product demo video is a 30-second script, real footage of your product, and an AI presenter who says the lines. Build it in that order and each AI step has a clear job, so you rerender less.

This guide covers the whole workflow, from the first line of the script to the version you post on each platform.

Pick the job before you write

Decide what the video must do. A teaser makes a stranger stop scrolling and click. A walkthrough shows a new user how to finish one task.

The teaser runs 15 to 30 seconds. It shows one feature and asks for one action. The walkthrough runs 60 to 90 seconds and follows one task from start to finish. If you try to do both in one cut, you get a long teaser that nobody finishes.

Write the script to a word count

Voiceover runs at roughly 140 words a minute. A 30-second teaser holds about 70 words. Count them. Cut until the count fits, not the other way round.

Give the teaser four beats:

1. Hook (first 3 seconds): the problem, in the words your customer uses.
2. Show (10 to 15 seconds): the product does the thing, on screen.
3. Proof (5 seconds): one number or one result.
4. Ask (3 seconds): one action, such as "Start free" or "Join the beta".

Write short clauses. A presenter, human or AI, reads short clauses cleanly. Long sentences with three commas come out flat.

Record the real product, not a generated one

Use a screen recording of your actual product for the "show" beat. Today's video models are good at people and light. They are poor at your exact interface, and a demo that shows a screen you didn't build misleads the buyer.

Record at a high resolution and zoom in during the edit. Small interface text turns to mush on a phone. Record one action per clip, so you can cut each one to the voiceover later.

AI video earns its place around the screen footage. Use it for the hook shot: a frustrated user at a desk, or a product photo that turns slowly. These shots don't need to show your interface, so the model can do what it does well.

Add a presenter for the parts that need a face

A presenter works best at the start and the end: a face that states the problem, then a face that asks for the click. Between those, cut to the screen. A talking head on screen for 30 seconds holds attention worse than one that comes and goes.

You need one portrait and one line of audio. Our post on turning a product photo into a talking spokesperson covers that technique step by step, and the talking avatar use case shows real results.

Some avatar models also speak the line for you. P-Video Avatar takes the script as text and has a choice of built-in voices in ten languages. You can also attach your own audio file, which replaces the built-in voice. At 720p it costs $0.04 a second, so two 5-second presenter shots cost $0.40.

If you use a real person's face, get their written permission first. That includes a cofounder.

Lock the voiceover before you cut

Make the voiceover first, then cut the pictures to fit it. The voice sets the timing. The presenter's lip movement follows the audio, and the screen clips follow the words.

Use one voice for the whole video. If the presenter speaks the hook and a narrator speaks the walkthrough, the change of voice sounds like an ad break. Keep music low and under the voice, or leave it out.

A clean voiceover also makes a second language cheap. Translate the script, generate the new audio, and run the presenter shots again. The screen footage stays the same.

Burn in the captions

Plan for viewers with the sound off. In a 2019 survey of 5,616 US adults by Verizon Media and Publicis Media, 69% said they watch video with the sound off in public.

Burn the captions into the picture, so they show on every platform and every player. Put them in the middle third of the frame, clear of the platform buttons at the bottom and the right edge. Check each word against the script. Speech-to-text often gets product names wrong.

Cut one version for each place it goes

Export from one edit, but change the shape and the length for each placement:

• TikTok, Reels and YouTube Shorts: vertical 9:16, 15 to 30 seconds. Put the hook in the first second.
• X and LinkedIn feeds: square 1:1 or 4:5, 30 to 45 seconds, captions on.
• Landing page and Product Hunt: horizontal 16:9, 30 to 60 seconds. It can loop with no sound.
• Help center and onboarding: horizontal 16:9, 60 to 90 seconds for one task.

A screen recording in 16:9 doesn't fit a 9:16 frame. Crop in on the part of the screen that changes, and put the presenter above or below it.

The takeaway

Write 70 words, record the real product, use an AI presenter only for the hook and the ask, and caption everything. Then export one cut for each place it goes.