Back to glossary

Term

Runway Gen-4.5

Runway Gen-4.5 is Runways video model that turns text or an image into 5- and 10-second clips with strong prompt adherence and cinematic motion. It was built with NVIDIA and uses an Autoregressive-to-Diffusion technique.

Runway Gen-4.5 — explained in more detail

Runway Gen-4.5 is the video generation model from the provider Runway and part of its creative platform and a developer API. It produces short clips via text-to-video and image-to-video with output lengths of 5 and 10 seconds. The focus lies on three areas: visual fidelity, motion quality, and prompt adherence. Objects move with realistic weight, momentum, and force; advances extend to fluid dynamics, surface rendering, and fine details such as hair and fabric textures.

Runway developed Gen-4.5 in collaboration with NVIDIA and uses an Autoregressive-to-Diffusion (A2D) technique running on NVIDIA GPUs. The model improves character and object consistency across frames, addressing a persistent problem in AI video generation where subjects drift or change appearance over the course of a clip. It maintains the speed and efficiency of its predecessor Gen-4. Gen-4.5 is a proprietary (closed) model with no open weights.

Example / In practice

A creative agency animates a storyboard still into a 10-second camera pan with believable physics and a stable main subject. Because Gen-4.5 keeps object consistency across frames, the character stays recognizable throughout the clip without additional rework.

Distinction from similar terms

Against Veo 3.1 and Kling 3.0, Gen-4.5 emphasizes physical fidelity, motion quality, and controllability rather than a native audio track or maximum clip length. In terms of purpose it belongs to image and video models and produces moving sequences with camera movement, not static single images.

See everything in one place:Image & Video