Luma Ray 2

Ray 2 is Luma AI's large-scale video generative model, built on a new multimodal video-transformer architecture with 10x the compute of its predecessor Ray 1. It turns text or image prompts into photorealistic clips up to 10 seconds long at up to 1080p, with natural coherent motion, physics-aware effects, and directorial camera control (pans, tilts, zooms, dolly shots) via natural language. It ships as the default model in Luma's Dream Machine platform and is also available through Amazon Bedrock and Replicate.

The honest picture: clips cap at around 10 seconds (extendable toward ~30s in some flows), there is no custom fine-tuning, and generation is tied to paid Dream Machine subscriptions with occasional slowdowns or failures at peak demand. At launch, image-to-video and video-to-video were announced as "coming soon" rather than shipped. It is a strong realism pick, but for native synchronized audio or longer single-pass narratives, rivals like Veo 3 and Seedance have pulled ahead on specific axes.