Google develops AI shopping tool that allows users to try on dresses and shirts at home

Last year, Google introduced an AI-powered shopping assistant in Search that allows users to get a visual idea of ​​how a particular outfit would look on a specific body type. Initially, the feature launched mostly with support for tops from a handful of women's clothing brands. Now, after more than a year of evaluating the benefits for both consumers and fashion brands, Google is expanding its AI shopping tool to include another clothing category: Dresses.

Shoppers in the US will now see a 'try on' badge appear when they search for a dress on Google. The AI ​​tool will show models in different body types, from XXS to XXXL. Users can select a model that fits their body type and try the dress on to see if it fits them before deciding to order. Once they find the right dress, they can click through to the retailer's website to buy it.

Google said it had to overcome many challenges to tweak the tool and also add support for dresses because dresses cover more of the body, and are "often more detailed than tops in terms of drape, style, length, or shape, and include everything from long camisoles to short skirts or off-the-shoulder maxi dresses and countless other styles."

To power this feature, Google couldn't just use the existing Virtual Try-On (VTO) AI model. While the VTO AI model is capable of diffusing low-resolution images, 'this approach often results in the loss of important details of the skirt.' To overcome this shortcoming, the Google team came up with a new training strategy called 'VTO-UNet Diffusion Transformer (VTO-UDiT).'

VTO-UDiT focuses on diffusing low-resolution images and gradually trains at higher resolutions during the diffusing process. This approach ensures accurate reproduction of fine details such as folds, patterns, and fabric textures.

Google currently has no plans to roll out this feature to other regions outside the US. It will take a lot more time to evaluate and finalize.