Luma | Ray3.2 Adds Frame-Level Control for AI Video Production
Luma has introduced Ray3.2, a major update to its AI video generation model focused on giving creators more precise control over motion, performances, framing, and production-ready output. The release combines frame-level direction with longer 1080p generations, enhanced performance tracking, HDR support, EXR export, and a new API for professional creative pipelines.
Ray3.2 brings frame-level direction to professional AI video
Ray3.2 moves the AI video workflow closer to traditional direction by allowing creators to control how a scene develops across individual moments. Instead of depending only on an initial prompt, production teams can establish important visual and narrative beats throughout a generated clip.
The model was developed with input from creatives working in entertainment, advertising, and gaming. Its new controls are intended to help artists manage pacing, camera movement, character performances, shot composition, and post-production requirements with greater precision.
Multi-keyframe control gives creators more precise direction
One of the main additions is multi-keyframe control, which supports up to 16 keyframes within a single clip. These keyframes can define changes in action, composition, camera paths, and visual progression, helping creators translate planned sequences or storyboards into generated video more accurately.
This approach can be useful when a scene requires specific timing or continuity. Advertising teams can coordinate product movement with planned story beats, filmmakers can guide a camera path, and visual creators can establish how a subject or environment should change between the beginning and end of a clip.
Performance tracking preserves gestures and facial expressions
Ray3.2 improves performance transfer through enhanced Performance Tracking and Expressive Facial Performance. The model can preserve skeletal posture and gestures while following the expressive state of up to eight faces simultaneously.
For character-focused video, this can help retain more detail from an acting reference instead of reducing the performance to general body movement. Facial expressions, posture, and gestures can remain connected across a scene, including shots with several visible characters.
HDR, EXR, Reframe, and longer clips support post-production
Native HDR generation and 16-bit EXR export make Ray3.2 more suitable for color grading and compositing pipelines. These output options give post-production teams more dynamic range when combining generated footage with visual effects, professionally captured material, or other design elements.
The enhanced Reframe feature can adapt aspect ratios, extend the visible frame, or replace a background while preserving the lighting of the original shot. Ray3.2 can also generate clips up to 20 seconds long at 1080p. Its complete control surface is now available through an API for custom creative applications, internal tools, and professional production systems.
Daisuki's Take: What This Means for Creative Workflows
Frame-level control addresses one of the most important limitations in AI video generation: the difficulty of directing how a scene changes over time. We see support for multiple keyframes as a meaningful step toward workflows where artists can plan a shot instead of repeatedly generating clips until the right movement appears.
The performance and post-production features are equally important. Facial tracking, gestures, HDR, and EXR output give editors and visual effects artists more opportunities to refine generated footage after it leaves the model.
Ray3.2 will still require careful testing for consistency, character behavior, and production requirements. However, its combination of direction, editable output, and longer clips shows how AI video tools are moving from isolated generation experiments toward more structured creative pipelines.
Sources and Recommended Links
- Luma Introduces Ray3.2 Model and API | Luma Official Blog
