Google announced on Thursday that it is expanding its generative AI-powered virtual try-on tool to include dresses, allowing users to virtually try on thousands of dresses from hundreds of brands, including Boden, Marge, Sandro, Simkhai and Staud.
The company says that dresses are one of the most searched apparel categories in its tool, but as Google explained in a blog post today, dresses are more detailed and complex than other clothing items, making them difficult to use with current virality technology.
To provide more context, Google Shopping last year released a tool that uses proprietary diffusion technology to create high-quality, lifelike images of tops and blouses. The tool simulates how clothing drapes, folds, clings, and forms wrinkles and shadows on real people in different poses.
Due to the intricacy of the dress's detail, existing diffusion models had a hard time accurately capturing the dress's fine prints, such as floral and geometric patterns. While the models could handle low-resolution images, the dress required a different approach to ensure important details were not lost. To address this issue, Google said it developed a new training strategy that starts with low-resolution images and gradually introduces higher resolutions.
Additionally, because dresses typically cover large parts of the body and come in a variety of lengths (midi, maxi, mini, etc.), placing a virtual dress on a person often obscures or blurs bodily details. A new technique called VTO-UNet Diffusion Transformer (VTO-UDiT) aims to solve this problem by preserving human features while erasing and replacing the dress, resulting in a more accurate depiction of both the dress and the person wearing it.
Virtual try-on technology aims to take the guesswork out of finding the perfect fit for customers of all body types. Various companies (Adobe, Amazon, Walmart) have launched their own tools to allow customers to virtually try on all kinds of clothing, including dresses. But with this new extension, Google seems to be trying to create a more advanced feature than its competitors.