Expect to see more AI-generated ads on the Amazon website going forward.
On Tuesday, Amazon announced the US release of its AI-powered video generator tool after nine months of beta testing. It also rolled out an updated version of the image generator first announced in 2023.
The video generator, which is now free to use for users with an authenticated Amazon seller account, allows for the creation of eight-second video ads (which the company describes as “low-motion,” presumably meaning that not much is actually moving in frame) from existing product images.
The tool can also take existing video assets, identify and summarize key clips and generate new videos from them for ad campaigns.
The AI models are trained on a variety of open, freely available data sources, including product information from Amazon Ads accounts.
Doing more with less
Most of the retailers who work with Amazon Ads lack the budget and expertise of professional marketers, Jay Richman, Amazon’s VP of product and technology, told AdExchanger.
Although some larger retail brands have embraced Amazon’s image and video generation tools, the main intention, Richman said, is to serve the needs of smaller, independently run businesses.
There’s been good adoption so far. According to Amazon Ads, 50% of the products promoted with AI-generated video assets are being advertised for the very first time, suggesting that these companies would not have previously been able to make these kinds of ads on their own.
Amazon wasn’t able to share whether AI-generated video ads contribute to higher sales for the businesses that use them. Richman did note, however, that sponsored brand campaigns featuring video elements see a 30% higher clickthrough rate on average.
‘Beyond white backgrounds’
The trend toward more video ads on Amazon has also given the company a larger catalogue of product pages and assets to train their generative AI tools on, allowing the technology to advance rapidly over the last few years.
When the original image generator was first launched 18 months ago, it was most useful for standalone products like chairs or appliances, but less so for products that “could not be easily placed on a flat surface,” Richman said.
For example, the previous generation of the tool had a hard time taking a photo of a wristwatch and making it look natural against an AI-generated backdrop. But now it can place the watch somewhere more intuitive, such as on the wrist of a photorealistic figure resembling a human model.
Similarly, the new video generator cana produce short animated scenes, meaning that same human figure can now adjust the watch on its wrist or the camera can zoom in on the watch as the figure drives a car.
Once these scenes are cut together and overlaid with written copy, which is separately generated by an LLM, the final result looks not unlike a professional (albeit slightly generic) video ad. Advertisers can also use the tool to modify the copy, add and adjust the placement of logos and edit the length and order of AI-generated clips.
“We’re not trying to recreate a full creative suite,” said Richman. “We want our customers who are not professional marketers to be able to get in and out and onto their day job, but to do so in a way that elevates their brand beyond white backgrounds.”
Product shown not actual size
Of course, anybody who’s accidentally bought a dud or a dupe on Amazon will have a natural follow-up question: What kind of guardrails does the company have to ensure these AI generation tools aren’t used to advertise fake items?
To that, Richman said the AI-generated ads go through the same moderation process as everything else on Amazon’s site. (Unfortunately, though, as both The Wall Street Journal and The Washington Post reported in 2019, Amazon’s automated moderation of third-party sellers frequently misses counterfeit and banned products.)
And there are also other limitations to the technology, at least for now. For instance, while the tool is more than capable of producing short, montage-style clips featuring the same or similar-looking human figures, it’s less adept at longer-form narratives.
However, Richman said he’s excited by how far the technology has already come in the last two years, and pointed out that this is a real product launch and far from the sort of “fantastical prototypes” that some AI companies rely on in their marketing pushes.
“This is like a real, practical application of generative AI that is live today and is ready to scale,” said Richman.
