Billions of people shop online every day, but replicating the in-store experience is challenging. Technology can help bridge this divide, but creating high-quality product visualizations can be costly and time-consuming. To address this, new generative AI techniques were developed to create shoppable 3D product visualizations from just a few product images. The latest advancement uses Google's state-of-the-art video generation model, Veo, to generate interactive 3D views for a wide range of product categories on Google Shopping. The first-generation approach used Neural Radiance Fields (NeRF) to render novel views, but suffered from noisy input signals and ambiguity from sparse input views. The second-generation approach used a view-conditioned diffusion prior to address these limitations, leading to significant scaling advantages and enabling the generation of 3D representations for many shoes on Google Shopping. The third-generation approach builds on Veo to generate 360° spins from one or more product images, generalizing effectively across a diverse set of product categories. This approach avoided the need to estimate precise poses from a sparse set of object-centric product images, increasing reliability. With as few as three images, Veo can generate high-fidelity and high-quality novel views, reducing hallucinations. The future outlook is to continue pushing boundaries to make online shopping more delightful, informative, and engaging for users.