What company's new generative AI model, capable of creating realistic 3D environments from text prompts, was unveiled in December 2025?

In December 2025, NVIDIA showcased significant advancements in its generative artificial intelligence models, particularly those capable of transforming text prompts into realistic 3D environments. Building on its expertise in graphics and AI, the company unveiled new capabilities within its "Cosmos" world foundation (Review) models. These models are designed to generate immersive 30-second predictive video worlds from various inputs, including text. The release of Image2Image and ImagePrompt features for Cosmos Transfer 2.5 on December 19, 2025, marked a notable step forward in enabling users to create detailed and controllable photorealistic simulations.

This innovation represents a leap in content creation, allowing developers and artists to rapidly prototype and populate virtual spaces with unprecedented ease. Instead of painstakingly designing every element, creators can simply describe their desired environment, and the AI model generates the corresponding 3D assets and scenes. This streamlining of workflows is crucial for industries that rely heavily on 3D content, such as game development, architectural visualization, and the burgeoning metaverse.

NVIDIA's ongoing commitment to generative AI extends beyond just creating static objects. These advanced models aim to build entire dynamic and interactive virtual worlds, accelerating the development of digital twins and sophisticated simulation environments. By providing powerful tools that translate natural language into complex 3D realities, NVIDIA is empowering a new era of digital creativity and efficiency, making the creation of rich virtual experiences more accessible and faster than ever before.