Share
Google just announced a new online shopping feature that leverages its new generative AI model developed by Google's shopping AI researchers. The company introduced a virtual try-on (VTO) feature for apparel, which allows users to preview clothes on real models with different body shapes and sizes. This provides a more accurate representation of how the garments would look in reality.
The objective of this AI model is to create lifelike portrayals of clothing on people. It takes into account essential details like draping, folding, clinging, stretching, and wrinkling. Unlike previous techniques that often resulted in visually defective and unrealistic images, Google's VTO feature generates every pixel of a garment from scratch, ensuring high-quality and realistic representations, the company claimed in a blog post.
To achieve this, Google's researchers used a diffusion-based AI model. They gradually add extra pixels or "noise" to an image until it becomes unrecognisable, and then remove the noise to reconstruct the original image in perfect quality. They modified the diffusion approach for VTO by utilizing a pair of images—a garment image and an image of a person. Each image is fed into its own neural network, known as a U-net, and through a process called "cross-attention," the networks share information to generate a photorealistic image of the person wearing the garment. This combination of image-based diffusion and cross-attention forms the basis of Google's new AI model for virtual try-on.
According to the company, to ensure realism and accuracy, Google extensively trained the AI model using its Shopping Graph database. They leveraged the comprehensive data set of products, sellers, brands, reviews, and inventory within the Shopping Graph. The model was trained with numerous pairs of images, each depicting a person wearing a garment in different poses. By learning to match the shape of the garment in one pose with the person in another pose, the AI model became proficient in generating realistic images of the clothing on individuals from various angles. This process was repeated millions of times using random image pairs of different garments and people, resulting in a powerful tool that allows users to see how a particular top would look on a model of their choice.
Starting today, Google's virtual try-on for apparel is available for women's tops from various brands across the Google Shopping Graph, including Anthropologie, LOFT, H&M, and Everlane. The tool will continue to improve and expand its reach to include more beloved brands, providing users with an unparalleled virtual shopping experience.
With Google's new generative AI model and virtual try-on feature, the company said that it believes shoppers can confidently explore and experiment with clothing options online. The company stated that the new feature will allow users to make more informed purchase decisions and bridge the gap between fashion's digital and physical realms.