Video generation AI 'Runway Gen-4.5' is now available, outperforming OpenAI's Sora 2 and Google's Veo 3



AI development company Runway announced its video generation AI, Runway Gen-4.5 , on December 1, 2025. Runway Gen-4.5 has improved prompt tracking and physical phenomenon reproduction capabilities, surpassing Google's Veo 3 and OpenAI's Sora 2.

Runway Research | Introducing Runway Gen-4.5

https://runwayml.com/research/introducing-runway-gen-4.5

Introducing Gen-4.5 | Runway - YouTube


The post below contains four videos that demonstrate the capabilities of Runway Gen-4.5. Runway Gen-4.5 is characterized by its excellent ability to understand prompts, and is able to correctly interpret and incorporate complex sequences of movements, detailed camerawork, event timing, and subtle changes in atmosphere into the video.




It also excels at physically accurate depictions, such as 'reflections in mirrors' and 'depicting a car shaking with the unevenness of the ground and its tires kicking up sand.' However, there are also problems that can occur, such as 'doors opening before the handle is pressed' or 'objects disappearing mid-video.'




Runway Gen-4.5's training and inference are performed on NVIDIA GPUs, and by working with NVIDIA, we have successfully optimized training efficiency and inference speed. Below is a graph showing the scores compared to 'Runway Gen-4.5,' 'Veo 3,' 'Veo 3.1,' 'Kling 2.5,' and 'Sora 2 Pro.' Runway Gen-4.5 reigns supreme, far ahead of other AI vendors.



Runway Gen-4.5 is being rolled out to a limited number of users in stages, with early partners including major American retailer Target, advertising firm Wieden+Kennedy, public broadcaster BBC, and game developer Ubisoft.



in AI,   Video, Posted by log1o_hf