Luma AI has shipped Ray3.2 — a video model built on a simple claim: creators shouldn’t just prompt, they should direct.
Frame-by-Frame Control
Ray3.2 allows users to direct AI video generation frame by frame, delivering cinematic-grade quality that integrates into professional pipelines. With the update, Luma says, “creators are no longer just prompting — they are directing.” It ships both in Dream Machine and as an API.
Scenes: Storyboard First, Render Second
Complementing Ray3.2, Luma Scenes structures video creation as a directing workflow — storyboard, key frames, and timing all reviewed before a single render. It’s the latest in Luma’s push from clip-generator toward production platform, following Layers (editable image layers), Pages, Skills, and the Uni-1.1 image API.
Open Physical AI Lab
Luma also announced its Open Physical AI Lab, an open-science effort to solve generalization in physical AI — arguing that AI systems must help improve the physical world, and that generalization is what stands in the way. The move parallels Runway’s world-models pivot, with both video companies racing to apply video-generation research to robotics.
Read the announcements at lumalabs.ai/news/introducing-ray-3-2 and lumalabs.ai/news/luma-open-physical-ai-lab.