Google has unveiled its latest breakthrough in the world of fashion & technology: a cutting-edge virtual try-on (VTO) feature for apparel. This innovative tool leverages artificial intelligence (AI) to provide users with realistic visualization of how clothes would look on models with diverse body shapes and sizes. It even simulates crucial elements such as draping, folding, clinging, stretching, and wrinkling. The generative AI model developed by Google’s shopping AI researchers is pushing the boundaries of virtual fashion. Let’s delve deeper into this exciting development and explore how it transforms how we shop for clothes online.
Revolutionizing Virtual Try-On: A Closer Look at Google’s Breakthrough
Google has disrupted the world of virtual try-on by introducing a feature that sets a new standard for realism. With this virtual try-on tool, users can now envision how clothing items would appear on real models, accurately representing various body shapes and sizes. By focusing on the essential details of how garments drape, fold, cling, stretch, and wrinkle, Google’s shopping AI researchers have developed a generative AI model that offers an unprecedented virtual shopping experience.
Also Read: Instacart Revolutionizes Shopping with AI-Powered Search: Meet Ask Instacart
Evolving Beyond “Clueless”: Overcoming the Limitations of Current Techniques
Virtual try-on has come a long way since its popular depiction in the movie “Clueless.” Previous techniques relied on geometric warping to fit clothing images onto silhouettes, often resulting in misshapen and unnatural appearances. These methods failed to adapt clothes to the body and often presented visual defects like misplaced folds. Google’s new virtual try-on feature aims to surpass these limitations by generating high-quality, realistic images of garments from scratch.
From Pixel to Perfection: The Power of Diffusion-based AI Models
To achieve unparalleled realism, Google’s researchers embraced diffusion-based AI models. Diffusion involves gradually adding extra pixels (or “noise”) to an image until it becomes unrecognizable and then removing the noise to reconstruct the original image flawlessly. By incorporating diffusion into their AI model, Google’s virtual try-on feature generates lifelike images of people wearing clothes.
Learn More: Join us for an extraordinary learning experience and unlock the boundless world of Generative AI with Diffusion Models at our upcoming workshop at the DataHack Summit 2023.
The Challenge of Visualizing Garment Fit: Combining Pose and Shape Adaptation
A critical challenge in virtual try-on is accurately visualizing how a garment will fit on an individual, considering significant pose and shape variations. Previous methods either focused on preserving garment details without accommodating pose and shape changes or allowed try-on with desired pose and shape but lacked garment details. Google’s solution combines advanced garment detail preservation with effective pose and shape variation, resulting in precise and realistic visualizations.
The Science Behind It: Understanding Diffusion in Text-to-Image Models
To grasp the inner workings of Google’s AI model, it’s essential to comprehend the concept of diffusion. In text-to-image models like Imagen, diffusion involves gradually adding noise to an image until it becomes unrecognizable and then removing the noise to reconstruct the original image. Combined with a large language model (LLM), this process generates realistic images based solely on text input.
Also Read: Nextech3D.ai Develops Generative AI That Converts Text to 3D Models
Unveiling a New Approach: Image-Based Diffusion with Cross-Attention
Taking inspiration from Imagen, Google’s researchers adopted diffusion for virtual try-on but with a twist. Instead of using text input, they incorporated a pair of images—a garment image and a person image—into the diffusion process. Both images interact with each other through a neural network, utilizing a technique called “cross-attention.” This novel combination of image-based diffusion and cross-attention forms the foundation of Google’s AI model for virtual try-on.
Harnessing the Power of Data: Google’s Shopping Graph Takes the Stage
To ensure the virtual try-on feature is as helpful and realistic as possible, Google’s researchers trained their AI model using the immense dataset available through Google’s Shopping Graph. This comprehensive data set encompasses the latest products, sellers, brands, reviews, and inventory, providing a wealth of information to enhance the accuracy and quality of the virtual try-on experience.
Also Read: SQL Powers to Reveal Insights into Brazilian Online Shopping
Mastering Realism: Rigorous Training with Millions of Image Pairs
Google’s AI model underwent rigorous training to master the art of generating realistic clothing visualizations. By using millions of image pairs that depicted people wearing garments in various poses, the AI model learned to match garment shapes with different body poses. This extensive training process empowers the AI model to generate highly realistic images of clothing on different models from multiple angles.
A New Era of Fashion Exploration: Virtual Try-On Available Now
Starting today, fashion enthusiasts can experience the virtual try-on feature for women’s tops from renowned brands across Google’s Shopping Graph. Brands such as Anthropologie, LOFT, H&M, and Everlane are part of this exciting launch. The virtual try-on feature promises to revolutionize the way users explore fashion choices online, providing an interactive and immersive experience.
Embarking on the Fashion Journey: Women’s Tops from Leading Brands
With Google’s virtual try-on feature, users can now effortlessly visualize women’s tops from a range of popular brands. Whether it’s exploring Anthropologie’s eclectic designs, LOFT’s versatile collection, H&M’s trendsetting fashion, or Everlane’s sustainable styles, users can now see how these garments look on models with different body shapes and sizes.
The Future of Virtual Try-On: Enhancements and Brand Expansions
Google’s virtual try-on feature is only the beginning. As the technology evolves and improves, the tool will become even more precise and expand to include a broader selection of users’ favorite brands. The future promises exciting enhancements & brand expansions, giving fashion enthusiasts the power to virtually try on clothes like never before.
Learn More: The DataHour: Data Science in Retail
Our Say
The line between online and offline shopping experiences blurs with Google’s virtual try-on feature. This remarkable innovation, powered by advanced AI models and a wealth of data, revolutionizes how we explore fashion choices. It also has the potential to influence how we make informed purchasing decisions. Virtual try-on will become increasingly realistic and accessible in the years to come. With such advancements, the future of online fashion is set to become more immersive, interactive, and tailored to individual preferences.
Sabreena Basheer is an architect-turned-writer who's passionate about documenting anything that interests her. She's currently exploring the world of AI and Data Science as a Content Manager at Analytics Vidhya.
