From product photo to video ad with AI: step by step

You make a video ad from product photos by animating them shot by shot with an image-to-video model, and then editing them together with a voice, text and your real logo. Generating is the easy part. The work lies in the first seconds, checking your product and a separate version for each platform.

On 12 June 2026, at the Google × ClickTrust event "Human in the loop" in Brussels, I built a video ad live from a few statics: still images of your product that you already have ^[1]. It's also how we approach AI video production at Izoard.

Step 1: gather your statics and write the hook first

A handful of varied images is enough: in the two cases on our studio page there were four and five, for ads of 18 and 16 seconds ^[1]. Make sure you have a high-resolution packshot, an image of the product in use, a detail and your logo as a separate file. Check the rights too: the licence of your photos must allow editing and ads, and anyone who animates a real person with AI needs that person's consent ^[2].

Then write the opening first. Google sums up its research into effective video ads as ABCD: attention, brand early on screen, human connection and a clear call to action. In a 2021 Google/Kantar study, such ads delivered up to a 30% higher likelihood of short-term sales ^[3].

Write three hooks right away: a different opening is just one new shot here, not a new shoot day. How to build them is covered in hooks for the first seconds of your video ad.

Step 2: animate shot by shot

For each shot, decide which static is the starting frame, what moves and what must not change. Veo 3.1 makes clips of 4, 6 or 8 seconds in 16:9 or 9:16, with sound ^[4]; Kling 3.0 goes up to 15 seconds ^[5]. In your prompt, describe the subject, the movement, one camera movement, the light and what stays fixed ("keep the label exactly as in the image").

Specify your packshot as the last frame, so the shot ends on your real product ^[4]. Ask for one thing per shot: the more you ask, the more the model fills in. And expect several takes per shot.

Step 3: check your product frame by frame

A generative model redraws your product frame by frame. Watch every chosen take slowed down on a large screen:

• Logo and text distort: place them as an overlay in the edit.
• Shape and colour shift: fix the first and last frame and correct with the static as a reference.
• Hands and liquid look unnatural: choose a shot without hands or a calmer camera.
• A promise that isn't true: cut the shot.

That last one is also a legal boundary. According to the European Commission, an AI image in which your product looks better than it really is counts as a deepfake; your real product against an AI background does not ^[6].

Step 4: add voice, text and music

Make the voice-over separately: Kling 3.0 doesn't list Dutch or French for its built-in audio ^[5]. A voice from ElevenLabs, for example, solves that, but only the paid plans include a commercial licence ^[7]. Plan a Dutch and a French version from the script stage, and use music with a licence that covers paid ads, which in our stack means Artlist ^[1].

Also put the core message on screen as text. Google recommends both audio and on-screen text ^[3], and says sound increases the conversions of Shorts ads by more than 20% ^[8].

Step 5: deliver per platform and label where needed

Generate vertically what runs vertically: Veo makes 9:16 directly ^[4]. Keep text and logo out of the interface zones: in Reels, 14% at the top, 35% at the bottom and 6% on the sides ^[9]. All specs are in video formats by platform.

Three hooks, two lengths and two formats give twelve variants of one concept. Meta explicitly built its ad system Andromeda for that growing number of AI creatives ^[10]; what that changes for your campaigns is explained by Jonas in Meta campaign structure in 2026.

Since 2 August 2026, realistic people who don't exist, or a real person doing something they never did, require a visible label. Your real product in motion or in an AI setting generally doesn't ^[6][2]. How to label is covered in Does your video ad need an AI label?

When are you better off filming?

Film, or combine, if you need to show exactly how your product works, want a real testimonial or need a long continuous shot. A generated shot can show something that isn't true, and Veo stops at 8 seconds ^[4]. That's why in the studio we mix AI with real footage where that is stronger ^[1].

Want to see what this does for your own product? Send your statics for a free test ad: we'll turn them into one video ad on your own brand, at no cost or obligation.