Google unveiled an update to its generative AI-powered virtual try-on tool, which now supports dresses, offering users the ability to virtually try on thousands of styles from numerous brands, including Maie, Boden, Sandro, Simkhai, and Staud.

Dresses have quickly become one of the most sought-after categories in this tool. However, Google acknowledged in a blog post that integrating dresses into their existing diffusion technique, has been challenging due to the garments’ intricate details and complex designs.

Initially launched last year, the virtual try-on tool utilized Google’s diffusion technology to create realistic images of tops and blouses, simulating how these garments would drape and fold. This technology struggled with dresses, particularly in accurately rendering detailed patterns such as floral or geometric designs. While the previous model managed lower-resolution images, it needed an upgrade to handle the finer details of dresses.

To address these issues, Google developed a new training strategy that starts with lower-resolution images and gradually incorporates higher resolutions to preserve detail. Dresses, which cover most of the body and come in various lengths like midi, maxi, and mini, posed the risk of obscuring or blurring body features.

To counter this, Google introduced a new method called the VTO-UNet Diffusion Transformer (VTO-UDiT), designed to maintain the wearer’s features while seamlessly integrating the virtual dress.

The aim of virtual try-on technology is to enhance the shopping experience by helping customers better visualize how clothing will fit, regardless of their body type. With competitors like Adobe, Amazon, and Walmart also offering virtual try-on solutions, Google’s expanded feature aims to deliver a more advanced and accurate experience.