#Video Generation

Ant Group open-sourced LingBot-Video, the world's first mixture-of-experts (MoE) based embodied video foundation model for robotics, achieving a state-of-the-art score of 0.620 on the RBench benchmark. The model is released under Apache 2.0 license on GitHub. This open-source release significantly lowers the barrier for embodied AI research, providing a highly efficient MoE architecture that activates only 3B of 30B total parameters, making it three times more efficient than dense models of similar size. It can accelerate progress in robot action prediction, simulation data generation, and world model development. LingBot-Video innovates in three aspects: architecture (DiT+MoE for capacity-cost balance), data (70K hours of embodied data covering dexterous manipulation, robot locomotion, and egocentric interaction), and training (multi-dimensional reinforcement learning rewards focusing on physical plausibility and task completion). The model uses a Diffusion Transformer (DiT) backbone.

Google has released Nano Banana 2 Lite, a fast image generation model that produces images in 4 seconds at a cost of $0.034 per 1K images, and Gemini Omni Flash, a multimodal video generation model that creates 10-second videos from text, image, or video inputs, now available to developers. These models significantly reduce the cost and latency of generative media, making high-quality image and video synthesis more accessible for developers and enterprises, and expanding the capabilities of Google's AI ecosystem. Nano Banana 2 Lite is available in Google AI Studio, Gemini API, and Gemini Enterprise Agent Platform, and will be integrated into consumer products like Search AI Mode and the Gemini app. Gemini Omni Flash costs $0.10 per second of video output but currently lacks audio reference, scene extension, and full character consistency across scenes.