Singapore-based startup Video Rebirth has secured $80 million in seed funding to advance its work on AI-driven world models, which are systems designed to simulate physical reality for applications in robotics, autonomous driving, and gaming. The company, founded in September 2024 by former Tencent AI researcher Liu Wei, focuses on generating lifelike video environments that strictly adhere to physical laws like gravity and collision. Video Rebirth currently operates with a 30-person team split between Singapore and Hong Kong.

In May, the startup’s flagship Bach model reached the No. 6 position on an Artificial Analysis text-to-video leaderboard, outperforming many larger competitors in cost-efficiency per generated minute of video. The model is capable of creating multi-shot videos up to 45 seconds long based on text or image prompts. Video Rebirth claims to achieve significant computational savings through a proprietary mathematical technique called multi-step sampling loss (a method for predicting and correcting generation errors to reduce required processing steps). This approach splits tasks between prompt adherence and visual generation, allowing the model to function with higher efficiency than traditional systems that rely on a single central brain for all operations.

The company aims to launch its advanced world model, named Olympus, by the end of 2026. This system is expected to generate not only 3D visuals but also environmental sounds in real time, enabling interactive simulation capabilities. Investors in the startup include AMD Ventures, Hyundai Motor Group’s ZER01NE, Actoz Soft, and Qiming Venture Partners. While competitors such as Google have introduced similar interactive technologies like Genie 3, Liu intends for the upcoming Olympus model to be a viable tool for industrial applications. Despite the high costs that led other major players to pause similar research projects, Video Rebirth remains focused on its goal of simulating the physical world through scalable and cost-effective AI architecture.

Singapore-based startup Video Rebirth has secured $80 million in seed funding to advance its work on AI-driven world models, which are systems designed to simulate physical reality for applications in robotics, autonomous driving, and gaming. The company, founded in September 2024 by former Tencent AI researcher Liu Wei, focuses on generating lifelike video environments that strictly adhere to physical laws like gravity and collision. Video Rebirth currently operates with a 30-person team split between Singapore and Hong Kong.

In May, the startup’s flagship Bach model reached the No. 6 position on an Artificial Analysis text-to-video leaderboard, outperforming many larger competitors in cost-efficiency per generated minute of video. The model is capable of creating multi-shot videos up to 45 seconds long based on text or image prompts. Video Rebirth claims to achieve significant computational savings through a proprietary mathematical technique called multi-step sampling loss (a method for predicting and correcting generation errors to reduce required processing steps). This approach splits tasks between prompt adherence and visual generation, allowing the model to function with higher efficiency than traditional systems that rely on a single central brain for all operations.

The company aims to launch its advanced world model, named Olympus, by the end of 2026. This system is expected to generate not only 3D visuals but also environmental sounds in real time, enabling interactive simulation capabilities. Investors in the startup include AMD Ventures, Hyundai Motor Group’s ZER01NE, Actoz Soft, and Qiming Venture Partners. While competitors such as Google have introduced similar interactive technologies like Genie 3, Liu intends for the upcoming Olympus model to be a viable tool for industrial applications. Despite the high costs that led other major players to pause similar research projects, Video Rebirth remains focused on its goal of simulating the physical world through scalable and cost-effective AI architecture.