Ray 2 is a large-scale video generation model developed by Luma Labs, officially launched in January 2025. Built on a proprietary multi-modal architecture, it was trained with ten times the computational power of the previous Ray 1 model. The model is designed to produce photorealistic video with a strong emphasis on natural motion, coherent physics, and logical event sequencing.
The model supports high-fidelity text-to-video and image-to-video generation. A key feature is its dual-image motion control, which allows users to set specific starting and ending frames to guide the animation trajectory. It also includes capabilities for seamless looping, cinematic camera movements, and high-resolution outputs ranging from 720p to 1080p, with 4K upscaling available.
Ray 2 is capable of generating base clips of 5 or 9 seconds, which can be extended to 30 seconds. Its architecture is optimized for production-ready content, featuring improved prompt adherence and the ability to render complex interactions between people, animals, and objects with minimal artifacts.