Google's new feature lets you virtually try thousands of dresses
Google has announced an expansion of its artificial intelligence (AI)-powered virtual try-on tool, now supporting dresses. This development allows users to virtually try on thousands of dresses from hundreds of brands including Maje, Boden, Sandro, Simkhai and Staud. The company revealed that dresses were among the most searched apparel categories for this tool. Google's feature expansion follows Adobe, Amazon, and Walmart launching their own virtual try-on technology, enabling customers to virtually try on various clothing items, including dresses.
Overcoming challenges with diffusion technique
Google acknowledged in a blog post that its current diffusion technique faced difficulties when applied to dresses due to their intricate details and complexity. The tool, launched last year by Google Shopping, uses this technology to generate high-res, lifelike images of tops and blouses. It mimics how the clothing would drape, fold, cling, or form wrinkles and shadows on real people in various poses.
New training strategy for detailed dress prints
The existing diffusion model had limitations in accurately capturing detailed dress prints like floral or geometric patterns. It could handle low-resolution images but struggled with dresses to avoid losing important details. To overcome this, Google created a new training strategy that begins with lower-resolution images and gradually incorporates higher resolutions.
Google introduces VTO-UNet Diffusion Transformer
Google has introduced a new technique known as the VTO-UNet Diffusion Transformer (VTO-UDiT) to address challenges posed by dresses that cover most of the body and come in various lengths. This method aims to preserve a person's features while erasing or replacing the dress, resulting in a more precise portrayal of both the dress and the person wearing it. The goal of virtual try-on technology is to eliminate guesswork when finding the right fit for customers of all body types.
