Powered by NVIDIA: CATCHES Launches Physics-Driven 'RealFit' for E-Commerce
Published: 23 Mar 2026
Author: Precedence Research
CATCHES leverages physics-based Artificial Intelligence to develop application and infrastructure layers for digital fashion. It launched 'RealFit', a Generative AI sizing technology on NVIDIA’s CUDA platform, designed to enhance online shopping with accurate sizing and fit.
Inside CATCHES’ RealFit: The Future of Digital Fabric and Size Accuracy
Unveiled at NVIDIA GTC, RealFit allows shoppers to visualize how garments will fit and move on their bodies with high precision. By entering their measurements and uploading a photo, customers create a digital twin to virtually try on items and compare sizes. The first public rollout of RealFit is now live on the AMIRI website, with additional brand collaborations and launches anticipated soon. CATCHES has raised $10 million in funding from prominent figures in the luxury sector, including Antoine Arnault, Natalia Vodianova Arnault, Roy Chung, Dillon Erb, Gary Sheinbaum, and Sarah Willersdorf.
According to Precedence Research, Applied AI in Retail and E-Commerce Market size accounted for USD 60.30 billion in 2025 and is predicted to increase from USD 72.42 billion in 2026 to approximately USD 376.48 billion by 2035, expanding at a CAGR of 20.10% from 2026 to 2035 as demand grows for hyper-personalization, predictive supply chain optimization, and generative AI-driven content creation to enhance customer experience.
CATCHES spent two years creating a unique simulation framework that uses GPU acceleration, designed and integrated with physical fabrics, before applying AI for scaling. The result is an accurate 1:1 representation of the actual fabric's weight, structure, drape, and movement. For consumers, RealFit offers the confidence of a fitting room experience from anywhere in the world. For brands, it provides a major chance to improve the connection between customers and products, increase purchase conversions, and significantly reduce fit-related returns, which are the main cause of reverse logistics costs in fashion e-commerce.
Physics-Backed AI: The Secret to Finding Your Perfect Online Fit
The virtual experience is built using NVIDIA CUDA and Omniverse libraries, utilizes the Newton Physics Engine, and combines custom-trained diffusion models with well-tuned VLMs and LLMs, including AI models based on NVIDIA Nemotron and NVIDIA Cosmos world foundation models, to speed up physically grounded generative experiences. CATCHES' bare-metal infrastructure is spread across NVIDIA RTX PRO™ 6000 Blackwell Server Edition, NVIDIA Hopper, and NVIDIA Blackwell and Blackwell Ultra GPUs, provided at scale with edge inference by Nscale, a vertically integrated AI hyperscaler, and through bare-metal partners Nebius, Boston Limited, and Scan Computers.
The system merges AI with NVIDIA CUDA and Omniverse-accelerated physics simulations to create digital twins with millimeter-level fit accuracy. By building the solution on an OpenUSD foundation, data can be shared easily between the digital twin model, Blender, and Unreal, while NVIDIA RTX Ray Tracing, DLSS, and DLAA guarantee photorealistic performance.
A recent report by Precedence Research highlights that the Applied AI in Retail and E-Commerce Market is benefiting from enhanced personalization, dynamic pricing optimization, AI-powered inventory management, and automated customer service, such as chatbots and virtual assistants.