How to make a product demo video with AI
An AI product demo video shows what a physical product looks like and how it is used, one feature per shot. The workflow is to list the three to five features that sell the product, plan one short shot per feature, build an accurate product reference set, generate each shot separately, and edit them with a real packshot at the end. The one firm rule: the video must not show the product doing anything it cannot do. This guide is for physical goods; software demos are still best recorded from the real screen.
Step 1: pick the features and the order
1. Write down every feature, then keep the three to five a buyer cares about most. For a travel mug: fits a car cup holder, leak-proof lid, keeps coffee hot, one-hand opening. For a backpack: laptop sleeve, hidden pocket, water-resistant fabric, compression straps.
2. Order them by what the buyer asks first, not by what engineering is proudest of. The opening feature should be the one that answers the main buying question in two seconds.
Step 2: plan one shot per feature
3. Write a one-line shot for each feature, naming the action that proves it. "Mug placed into a car cup holder, close-up." "Mug turned upside down over a laptop keyboard, nothing drips." "Thumb flips the lid open while the other hand holds a phone." A demo shot shows; it does not tell.
4. Choose a shot size per feature. Close-ups for mechanisms such as lids, zips and buttons. Medium shots for use in context, such as the bag on a shoulder in a train station. One wide shot is plenty.
5. Keep each shot short. Four to eight seconds per feature is enough. Kling 3.0 Omni generates up to 15 seconds per request, according to Kling's user guide as of October 2026, and RefVid clips run 3 to 15 seconds, but a 30-second demo built from six short shots is easier to watch and to fix than two long ones.
A typical 30-second plan: a 3-second opening hero shot, four feature shots of about 5 seconds, and a 4-second packshot with the name and price.
Step 3: references that show the mechanisms
6. Build the usual product set, hero angle, front, side, back and a logo close-up, and add images of the parts the demo depends on: the lid open and closed, the laptop sleeve unzipped, the hidden pocket with the flap lifted. A video model cannot invent a mechanism correctly if the references never show it; it will invent a plausible but different one.
Some models also take a short real clip, such as a lid opening: Kling 3.0 Omni accepts one reference video of 3 to 10 seconds with up to four images, according to Kling's user guide as of October 2026.
RefVid takes 1 to 5 photos per product and no reference video, so photograph each state as a still. If the mechanism shots need more than five images, save a second subject for that state, such as "Mug, lid open", and use it for those shots. Product photos can also be imported from your shop's product page link.
Step 4: generate and check each shot
7. Write each prompt around the action and the end frame. "Close-up on a kitchen counter. A hand presses the button on the lid and it flips open; steam rises. Static camera, soft daylight from the left. Ends with the open lid facing camera." Leave the product's appearance to the references.
8. Check every shot against the real product, not against your memory of it. Does the lid hinge on the correct side? Is the number of straps right? Does the zip run the right way? Mechanism errors are easy to miss in motion; scrub through frame by frame.
9. Regenerate shots that show a feature wrongly. Never keep a shot where the product does more than it does in reality, such as a mug that stays visibly steaming for hours or a bag that repels a downpour the fabric is not rated for. That is a misleading ad, not a creative choice.
Step 5: edit and finish
10. Assemble the shots in your editor in the planned order, add short on-screen labels for each feature in large type, and add music or voice. End with a packshot made from a real product photo or 3D render with the correct logo, name and price, since printed text in generated video is not reliable.
Make a 16:9 version for the product page and YouTube and a 9:16 version for social. Generate each orientation natively rather than cropping, so the product stays centered; RefVid generates 16:9, 9:16 and 1:1.
RefVid saves the product once with its photos and lists every demo shot on that product's page, which makes it quick to compare shots for consistency. Editing and on-screen text happen in your own editor.
Common mistakes and checklist
Mistakes: features described in voiceover but never shown; mechanisms missing from the references; one long shot that tries to cover everything; exaggerated performance; label text taken from the generated video.
Checklist: three to five features chosen and ordered; one shot per feature with a provable action; mechanism images in the references; shots of four to eight seconds; each shot checked frame by frame against the real product; real packshot at the end; 16:9 and 9:16 versions.
Frequently asked questions
- Can AI make a product demo video without filming the product?
- Yes, if you have good photos of the product, including its mechanisms. The video model rebuilds the product from those reference photos, so anything not shown in them may be invented.
- How long should a product demo video be?
- Around 30 seconds for social and product pages: a 3-second opening hero shot, three to five feature shots of about 5 seconds each, and a packshot of about 4 seconds.
- Is it legal to show AI-generated product footage in ads?
- Generally yes, as long as the AI footage does not mislead buyers about what the product does. Check every shot against the real product and avoid exaggerated results. This is not legal advice.