Virtual try-on in retail

Virtual try-on now allows customers to try on clothing, shoes, or accessories digitally – without delivery or a fitting room. Based on AR, AI, and computer vision, its potential to reduce returns and increase conversion rates is widely discussed. But is that truly realistic?

Virtual try-on works in three main steps: capturing the body, modeling the product, and rendering it realistically.

1. Capture: First, the smartphone camera analyzes body contours, poses, and proportions in real time using computer vision algorithms such as MediaPipe. The result is a digital approximation: keypoints plus silhouette.
2. Modeling: At the same time, the product, such as a garment, is prepared as a 3D model with textures, fabric simulation, and lighting adjustments, often using CAD tools such as CLO3D or Style3D, or photogrammetry.
3. Integration: Finally, AR software such as ARKit or ARCore merges the two into a photorealistic overlay that dynamically reproduces movement and lighting conditions. Retailers provide API-based 3D product models so apps or online stores can load the visualization almost instantly. Processing is often not handled entirely locally, but through hybrid cloud-edge architectures or fully server-side.

Systems such as Perfect Corp.’s YouCam further optimize this process with generative AI, which adapts clothing seamlessly to individual body shapes and provides retailers with precise fit feedback.

Virtual try-on is especially relevant in retail sectors where fit, size, or visual effect are difficult to judge before purchase. In fashion in particular, it can help reduce uncertainty around size, style, and fit. In beauty and eyewear, the technology is especially suitable for makeup and glasses because customers can quickly compare colors and shapes visually. Shoes and accessories can also be assessed more effectively, especially in terms of proportions, overall effect, and size relationships. Beyond that, virtual try-on can also make sense in furniture and home retail, for example to test spatial impact and dimensions in advance.

• Fewer returns: Returns can be avoided because customers can better assess fit beforehand.
• Higher conversions: More purchase confidence through realistic visualization and longer dwell time.
• Better customer insights: Fit data can optimize product development and assortment planning.

• Technical barriers: 3D models and API integration
• Data protection: biometrics under the GDPR, and bias in relation to diversity
• Accuracy: lighting and fabrics can interfere with rendering

The Wearfits study “Why Shoppers Admire Gen AI Virtual Try-On But Rarely Buy” points to a striking paradox: although, according to the study, 77 percent of online shoppers want virtual try-on, actual use remains low. The return rate for fashion items is 25 percent, well above the e-commerce average of 16.9 percent. While the technology is technically impressive, with physics-based fabric simulation and generative models, it still runs into trust barriers. Customers hesitate to upload personal body scans because they fear data breaches – 42 percent hesitation according to the study – and because of the uncanny valley effect, where avatars appear unnatural.

In addition, a promise-reality gap makes the problem worse. Advertising promises perfect personalization, but what customers often see are generic models with inaccuracies in patterns, logos, or prints caused by pixel distortion. Solid colors often work better. This leads to frustration, abandoned purchases, and blame directed at brands. In its study, Wearfits argues for transparency: limitations should be disclosed, for example through EU AI Act labels; “good enough” accuracy should be the goal; and trust should be built step by step. Perfection is an illusion – honesty is the path to habit.