New Runway Model Enhances Physics and Detail in AI Video

In the generative AI space, where competition intensifies monthly, an update has arrived tackling one of the most challenging tasks—video creation. Runway announces the release of its Gen-4.5 model, which, according to the company, can produce video clips with a heightened level of realism.

The developers' primary focus in this version was on visualization accuracy and the plausibility of physical processes. The new system better handles complex user prompts and strives to avoid contradictions in scene logic. Improvements have been made to the depiction of object motion, their interactions, and the behavior of various materials.

Despite the stated improvements, the technology still faces certain challenges. Occasionally, generated clips may exhibit breaks in cause-and-effect relationships; for example, an action might occur before the event that triggered it. Active development of video generation tools is not limited to Runway. Other major players, like OpenAI with its Sora 2 model, are also working on enhancing realism and accuracy in reproducing complex scenes, including stunts and interactions with water surfaces.

The Gen-4.5 model is gradually becoming available to users of the Runway platform. The company states that request processing speed will remain at the level of the previous version, ensuring no increase in wait times for results despite the increased computational complexity.

Progress in video generation is gradually blurring the line between AI-generated and camera-captured content, which in the long run could transform the approach to producing visual materials for various industries.