PANews reported on October 16 that WorldLabs, a spatial intelligence company founded by Stanford University professor Fei-Fei Li, officially released a research preview version of its new generation real-time generation model RTFM (Real-Time Frame Model), which can generate videos and 3D scenes in real time during the interaction process, realizing a persistent virtual world. RTFM is designed around three core principles: efficiency, scalability, and persistence. The model runs at interactive frame rates on a single NVIDIA H100 GPU, generating scenes with complex effects like reflections, shadows, and glossiness from a single image without explicit 3D modeling. Its "spatial memory" mechanism maintains world persistence through frame pose modeling and context juggling, supporting long-term interactive exploration.PANews reported on October 16 that WorldLabs, a spatial intelligence company founded by Stanford University professor Fei-Fei Li, officially released a research preview version of its new generation real-time generation model RTFM (Real-Time Frame Model), which can generate videos and 3D scenes in real time during the interaction process, realizing a persistent virtual world. RTFM is designed around three core principles: efficiency, scalability, and persistence. The model runs at interactive frame rates on a single NVIDIA H100 GPU, generating scenes with complex effects like reflections, shadows, and glossiness from a single image without explicit 3D modeling. Its "spatial memory" mechanism maintains world persistence through frame pose modeling and context juggling, supporting long-term interactive exploration.