Ray 3 is a high-fidelity video generation model developed by Luma Labs, representing a significant advancement in generative cinematography and reasoning-driven video synthesis. Positioned as a "reasoning" model, it is designed to interpret complex prompts with a deep understanding of semantics, physics, and scene logic. This capability allows the model to plan coherent sequences, maintain character consistency across shots, and self-correct during the generation process to produce more logical and physically accurate motion compared to previous iterations.
A defining technical feature of Ray 3 is its support for native High Dynamic Range (HDR), specifically 16-bit ACEScg. This provides professional-grade contrast and color depth, making it suitable for studio-level post-production workflows. The model supports text-to-video, image-to-video, and video-to-video tasks, delivering cinematic textures and sophisticated camera movements through a multimodal architecture that scales compute significantly beyond the earlier Ray architectures.
Ray 3 introduces advanced creative controls, including an annotation tool that enables users to guide layout and motion by drawing directly on input frames. It also features a specialized "Character Reference" system to ensure visual continuity in multi-shot narratives and a "Draft Mode" for rapid experimentation at higher speeds. These features are aimed at reducing the iteration cycles typically required for professional video production using generative AI.