The experiment
Nykaa Fashion had thousands of products competing for attention, but producing dedicated video for every SKU was expensive and operationally heavy.
We ran a small pilot around a simpler question: could the product images Nykaa already had become usable video without organizing another shoot?
For the experiment we received a batch of 10 lower-performing SKUs from the Nykaa store. Each product already had roughly three to four existing product images. There were no new shoots, no models to coordinate, and no additional photography. The constraint was the point: work entirely with assets that already existed.
What we built
This was an earlier generation of AI video, when image-to-video models were far less consistent than they are now. We did not expect a model to produce a finished advertisement in one shot. We built a small production workflow around the limitations.
We used Runway and FloraFauna AI to take the existing photographs and generate short motion sequences from them. The path was roughly: existing SKU images → AI-generated motion → select usable generations → edit sequences together → add sound → export a finished product video.
The AI was only one part of the system. Some generations distorted the product, changed details, or produced motion that was not usable. The work was creating a repeatable process for generating multiple candidates, picking the strongest clips, and assembling them into something commercially usable.
For every SKU, the original photographs became the raw material for a short product video. We then stitched the clips in a traditional editor, added pacing and sound, and produced the final assets.
One asset, multiple surfaces
The more interesting outcome was not simply putting video on a product page. Once a SKU had been converted from still photography into motion, the same underlying asset could be adapted for product detail pages, Instagram and other social platforms, paid advertising, retargeting, marketplace listings, email, and short-form promotional cuts.
Instead of treating AI video as another piece of content to produce, the experiment suggested a different model: existing commerce photography could become the source material for an entire layer of video. A retailer would not necessarily need another shoot every time it wanted motion.
What happened
We were told the experiment resulted in higher impressions for the selected products. We do not have sufficiently verified analytics from the pilot to claim a specific uplift in impressions, engagement, or conversion, so we do not present those numbers as measured results.
The pilot was more valuable to us as a systems experiment than as a performance-marketing case study. It showed that existing product photography could be turned into video through a repeatable, AI-assisted production pipeline.
Why we didn’t continue
There was also a commercial lesson. At the time, video generation was relatively expensive and unreliable. Producing enough generations to get clean, usable clips took significant compute, experimentation, and manual editing.
For a batch of only 10 products, that was manageable. At catalog scale, the economics were different. The cost of generation left too little margin for us to operate the workflow profitably as a service, so we chose not to pursue it further.
That decision belongs in the note. The technology worked well enough to demonstrate the system. The unit economics were not yet attractive enough to turn it into a scalable NextGrid offering.
What we learned
The experiment changed the question from “can AI generate a product video?” to “can an existing product catalog automatically become a video catalog?”
For those 10 Nykaa SKUs, the basic pipeline was possible. The bottleneck was not access to creative assets — Nykaa already had them. It was not necessarily production either. Much of that could be automated. The bottleneck was the economics and reliability of the video models available at the time.
As those constraints improve, the same workflow becomes more interesting: take photography that already exists across an ecommerce catalog and systematically turn it into reusable motion for the site, social, advertising, marketplaces, and whatever channel comes next.
Questions
Was this a production catalog rollout?
No. It was a 10-SKU pilot using photography Nykaa already had. The pipeline worked well enough to demonstrate the system. The unit economics were not attractive enough for us to operate it as a scalable service, so we did not pursue it further.
What was the real bottleneck?
Not access to creative assets. Nykaa already had them. The bottleneck was the economics and reliability of the video models available at the time. As those constraints improve, the same workflow gets more interesting.
