Luma has released Ray3.2, an updated version of its AI video generation model developed with input from creatives in entertainment, advertising and gaming.
The update focuses on tighter control over generated video, including frame-level direction through multiple keyframes within a single clip. Users can place up to 16 keyframes to shape motion, pacing and scene progression.
The release marks a push toward professional production work rather than consumer experimentation. It also expands Ray beyond a standalone tool by making the model's full control surface available through an API for developers.
The API allows integration into studio software, in-house production tools and other creative platforms. The aim is to let teams use the system within existing workflows instead of relying on separate interfaces or manual handoffs.
Ray3.2 also supports HDR generation and 16-bit EXR export. Those formats are widely used in visual effects, compositing and colour workflows, where teams need image data that fits established post-production processes.
Luma also said the updated model can preserve more of an actor's movement and facial expression during generation. Performance tracking now covers skeletal posture and gestures, while facial performance can follow up to eight faces frame by frame.
Another feature targets post-production changes after a shot has been generated. The enhanced reframe tool can alter aspect ratios, extend a frame or replace a background while maintaining the original lighting conditions.
Ray3.2 can generate clips of up to 20 seconds at 1080p. That is intended to support longer continuous scenes rather than short snippets, an area where many generative video tools remain limited.
Studio focus
The emphasis on directability reflects a broader shift in the AI video market, as suppliers try to appeal to film, advertising and game production teams that need consistency and editability as much as image quality. Tools aimed at those buyers are increasingly judged on how well they fit established pipelines and handle revision cycles without forcing a scene to be rebuilt from scratch.
Luma framed the release around that production need, particularly for agencies and studios working to detailed briefs and storyboards. The combination of keyframing, performance retention and export formats suggests a strategy aimed at professional users who need more predictable outputs.
Developer route
The API launch could also widen Luma's reach beyond direct users. By exposing the same controls through software interfaces, the company is giving developers a way to embed video generation into proprietary systems, render workflows and third-party creative products.
That matters in production settings, where teams often rely on connected software stacks rather than individual tools. An API approach can make it easier for media businesses to test generative systems within existing asset management, editing and review environments.
Competition in generative video has intensified as providers try to differentiate on controllability, output quality and workflow integration. While consumer-facing, prompt-based video systems drew early attention, vendors are now under pressure to show that their models can meet the technical and operational demands of commercial production.
Luma said Ray3.2 was built to address that requirement through more exact direction and outputs intended for post-production use. "With Ray3.2, direction goes in, and cinema comes out," Luma said.