How to make an AI product video

What is an AI product video?

An AI product video is a short ad where the product is the subject of every shot: a hero close-up, the product in use, a detail of the material, and an end card. AI generates the scenes, light and motion; the product itself comes from a real photo.

It differs from a UGC ad in who carries it. In a UGC ad a person talks; in a product video the product and the camera do the work, with a hand or a presenter at most. Public examples of AI ads in this style are on AI product demo ads and AI product placement ads. Those come from posts on X, not from Anglr.

Tools creators use for it

The counts show how many of the reviewed product video tutorials used each tool.

Based on 31 public tutorials posted between 15 Jan 2026 and 10 Oct 2026, reviewed by Anglr.

How do you make an AI product video step by step?

You make it in six steps: packshot, storyboard, stills, motion, edit and test. This is our synthesis of the tutorials, with the points where they agree and differ.

1. Get a clean packshot of the real product

Photograph the product, remove the background and make sure the label is sharp and readable. Mayur jadav starts from a phone photo he cleans up. Muhammad Wahaj treats label accuracy in the source photo as the first check. Estefania Freites and Nipun Taneja also start from one product photo.

For multi-angle shots, Estefania Freites builds a product sheet with front and three-quarter views so the video model has the sides it would otherwise invent.

2. Plan the shots as ad beats

Plan six to ten shots before you generate anything, and give each shot a job in the ad. At least five authors map shots to beats. A common order is:

Estefania Freites uses six beats close to these. Muhammad Wahaj plans ten scenes from a hook close-up to a hero shot. Abu kalam khandaker ends on a product lineup with a call to action. Dusan Radovanovic maps a 15-second ad into beats before writing any prompt.

3. Draw the storyboard as one image

Several creators generate the whole storyboard as a single image: a grid of frames, one per shot, with the product in every frame. @Strength04_X uses an eight-scene page in Nano Banana 2 with a rule that the product is the hero of every scene. @AiwithLariab uses a ten-frame grid in GPT Image 2. Israr Ahmad and Sami Hasnain plan scene by scene in the same way.

The board does two things: it lets you check shot order and product consistency before spending video credits, and it gives the video model one reference for the whole ad.

4. Make each still, then animate it

Generate a still for each shot with the packshot attached, then animate the still. This two-step split, image first and motion second, is described by @ShamiWeb3, @oggii_0, Israr Ahmad and Abu kalam khandaker.

A related point comes up across several sources: do not ask one prompt for the finished ad. @AmControo splits colourways and motion across two tools. @piyascode9 builds a white 3D model as the base first. @shabnam_774 takes the product through a clay stage. Dusan Radovanovic separates planning in Claude from rendering in Seedance.

Where sources differ is the motion model. Some animate in Seedance, some in Veo inside Google Flow, some in Kling. The Google Flow users (Muhammad Wahaj, Mayur jadav, Sami Hasnain) upload the product as a reference so the label stays the same.

5. Describe materials and light exactly

Name the material and the light in every prompt: brushed aluminium, frosted glass, matte card, condensation, steam. Dusan Radovanovic names materials so texture reads as real. Israr Ahmad builds food shots on macro texture. @MrOnsase uses macro and slow-motion detail for a perfume. This is where AI product photography and product video overlap: the still has to look like a real studio shot before it moves.

6. Edit, then test variants

Cut the shots short, add type, sound effects and the end card. Abu kalam khandaker cuts every three seconds. Alfredo Jair San Juan Hernandez cuts between presenter, close-ups and product shots. Estefania Freites and @AmControo finish in CapCut.

Then test several versions. Nipun Taneja checks that the hook works and that the ad reads with sound off before launch, then runs four to six variants. Estefania Freites uses each beat as a standalone test creative as well as part of the full ad. @darshal_ frames the goal as a library of content from one product rather than one ad.

Single-author methods worth knowing

• @SaniBulaAI recommends an exploded-view shot: generate the assembled product as the first frame and the separated parts as the last frame, then animate between them. @Strength04_X uses a similar hero-then-deconstructed pair.
• @TechWith_Nova edits existing product footage with Runway Aleph instead of generating new footage.
• Deykhan Ten filters weak and duplicate product photos before generating, and renders several aspect ratios at once.

Prompt templates for a product video

These templates are ours. The still follows Google's documented Nano Banana product structure (product on a surface, lighting and its purpose, camera angle, the detail in focus, then aspect ratio). The motion prompt follows Google Cloud's Veo order: camera, subject, action, context, style, then sound. More examples are on Nano Banana prompts and Veo 3 prompts, and you can build your own in the AI ad prompt builder.

Hero still (Nano Banana, product photo attached):

Use the attached photo of [product] exactly as it is: same shape, colours and label. A studio-lit product photograph for a [category] ad. The [product] stands on [surface] beside [one or two props]. Lighting: [key light] from the left to show [material], soft fill from the right. Camera: low angle, macro lens, sharp focus on the label and [detail]. Clean background in [colour]. 9:16.

Motion (Veo, the still as the reference image):

Slow push-in, macro lens. The [product] from the reference image stays still on [surface] while [motion: water beads run down the glass / powder lifts into the air / the lid turns open]. Context: [setting], [time of day]. Style: premium product commercial, shallow depth of field, crisp highlights. The product shape, colour and label stay exactly as in the reference.
SFX: [sound of the action].
Ambient noise: quiet room.

Common problems and fixes

What creators report

These are the authors' own claims.

• Muhammad Wahaj reports making a beverage commercial for a client in 23 minutes.
• Dusan Radovanovic reports making eight shoe ads in one afternoon.
• Nipun Taneja reports making a video ad in 11 minutes, down from 6 days.

Anglr pays AI creators and editors on performance to make product ads like these from winning scripts; see Anglr for creators for what we look for.