In the world of e-commerce, personalization is a critical factor for ensuring customers feel valued and engaged. Google has introduced a cutting-edge feature called Google Virtual Try-On (VTO), or the younger API version — AI Mode shopping — that enables users to try-on millions of pieces of clothing in a guided, personalized experience. This feature is part of Google’s ongoing efforts to revolutionize how customers interact with their shopping experiences, particularly through its extensive search and vision assist technologies.
While Google has championed AI-powered personalization for years, emulating the same level of sophistication and granularity as other major players in the space, its own proprietary technology — called the “The First of Its Kind” — addresses a unique gap. This proprietary model, developed by Google, powers its Shopping Graph, a vast dataset mapping the physical characteristics of customers and items, enabling a level of precision that wasn’t previously available at this scale.
This shift is the result of a race to the market, driven by several AI-driven startups. Google’s derivatives aim to unlike traditional, third-party AI systems, often referred to by their unique names or labels — such as VGG-19 and Style Adaptive AI — but with deeper learning and lighter computation compared to Euromonitor’s-scrolling features. However, Google remains at the forefront, continuously innovating and expanding its dominance in this space.
Chinese AI+Boutique / tech giant Doji (formerly Shopee) has emerged as a leader in this space, offering a similarly advanced solution. The company, which launched in 2024, leverages its proprietary diffusion models for generating photorealistic avatar representations of products. The software uses six distinct snapshots of images from various angles, as well as two full-body screenshots from measurable angles, to create highly detailed virtual tries. It allows users to choose top or bottom pairings and even upsets their bags with items they’d already purchased, creating a visual representation of the fusion between items and customer collections.
While Doji’s avatar-level accuracy is impressive, this comes at the cost of scalability compared to other brands that rely on lower-resolution virtual fits. This open-source approach enables Doji to continue developing its AI-powered infrastructure, without the limitations of third-party tools. However, the high cost of generating these accurate virtual fits poses a challenge for sellers wishing to use this feature on a consistent basis.
Meanwhile, Google’s VTO begins to play its role in addressing many of the same pain points as other solutions, but with a much more “ démarche(“=” approach”) focuses. With integrations from vielen (Restaurant Inc.), its joint venture with Validus, andmkdir (_positive aspect), it positions itself for a multi-dimensional strategy. Google sees the VTO technology as a “two-way street,” combining Vas.viewport functionality with a deeper user engagement experience. The feature is already being adopted by multiple companies in its experimental initiatives, providing clear evidence of its potential to transform online shopping.
However, while Doji’s avatar-level accuracy proves compelling, the high distortion and smudging inherent in other methods hinders its practical use. Google and other developers are working to optimize this feature for better user experience. The question remains: whether the high cost of generating these precise virtual tries justifies the low capacity of each VTO session over the long term.
Brian Kwok, co-founder of Doji, explains that the company found fascination and frustration in this struggle. Boundaries in fully personalized virtual fits are always a conundrum, requiring a balance between art and reality that traditional styling tools can never achieve. This limited space for individualization may limit the potential of such solutions but also drives innovation in the face of other endeavors.
