Google's Gemini Omni 1.2 Flash, a model that makes video from text, photos and short reference clips, now produces clips up to 15 seconds long, up from 10. You can ask for any length from 3 to 15 seconds. It costs 30% less than the version before it, at $0.07 per second for 720p video, and it keeps faces, clothes and packaging looking the same as the camera moves around them.
Editing by conversation was already part of the earlier version. What changed is that it now works across the full 15 seconds, and that each try costs much less.
Cheap drafts make a whole catalog affordable
Google's own example gives the idea. One photo of a model, plus front and back photos of three outfits, produces a 15-second try-on clip with the model turning 360 degrees and every garment keeping its cut and drape.
Now take a small online shop with 40 products. It can draft each clip at the lowest quality, 360p, which costs about 2 cents a second, so roughly 35 cents for a 15-second draft. Ten drafts of one product come to about $3.50. Add a dollar or so for the final version at 720p, and each product costs about $4.50. The whole catalog is under $200, which is a fraction of what one day with a photographer, a model and a studio costs.
The same clip can be sold many ways
The second change is how cheaply one clip becomes many. Google's example starts with a single scene of two characters and swaps the background turn by turn, from a tent to a forest, a stable, a market fair and a castle, while the characters and the camera stay the same.
For the shop, that means the same jacket spin in snow for one audience and on a city street for another, vertical for stories and square for a feed. Each version is one follow-up message, with no new brief and no new shoot. Google has also put the model into its ad tools, so the loop of making a version, running it and making the next one can happen in the place where the ad is bought.
My forecast is that product video becomes as ordinary on a listing as product photos are now, and that a listing without it starts to look unfinished. Photos went that way when taking them became cheap, and the shops that gained most were the ones that learned which photo sold.
Where it still falls short, and who can use it
Google names fast sports action, quick changes of script and complex physics as things it is still improving. Slow product turns, a person talking to the camera and a smooth camera move through a building are what the model is good at today. A ball game is not.
Developers and businesses can use the model through the Gemini API and Google Cloud. In Google Vids, anyone with a Google account can make 15-second clips at no extra cost. In the Gemini app it is rolling out first to subscribers on the most expensive plan, Google AI Ultra. Every clip carries an invisible watermark that marks it as AI-made.
The one check that stays with the owner is whether the clip shows the jacket the way it really looks. A customer who receives something different sends it back, and a return costs far more than a dollar.