Ray 1 is the foundational large-scale video generation model developed by Luma Labs. Formally released in June 2024 as the core engine for the Dream Machine platform, it was one of the first high-quality video models made available to the general public. It is designed to generate realistic, high-fidelity video clips from natural language descriptions and static images.
The model is built on a multimodal transformer architecture and was trained directly on video data. This allows Ray 1 to function as a "world model," providing it with an inherent understanding of physical interactions, motion coherence, and scene logic. At its launch, the model was capable of generating 5-second video sequences (120 frames) with cinematic camera work and consistent lighting, supporting a wide range of aspect ratios and resolutions up to 720p.
Capabilities and Features
Ray 1 supports both text-to-video and image-to-video workflows. In image-to-video mode, the model can interpret the context of a static image to animate it realistically, maintaining character consistency and environmental details. The model's training focuses on temporal consistency, ensuring that movements—such as water flowing or characters walking—remain fluid and physically plausible throughout the duration of the clip.