Luma Ray 3.2 pairs cinematic HDR generation with frame-level keyframe control, letting you choreograph exactly where motion lands, camera moves, and beats hit. Released June 2026 for text, image, and video-driven production.
Place up to 16 individual keyframes inside a single clip to decide exactly where the look shifts, a movement lands, and a beat hits.
Generates true high-dynamic-range color natively, with EXR export in ACES2065-1 (AP0) for professional compositing pipelines.
Produces native 1080p output stretching to 20 seconds for modify workflows, across 360p, 540p, 720p, and 1080p.
Preserves the full expressive state, lip sync, and dialogue timing for up to eight faces simultaneously, frame by frame.
Luma Ray 3.2 is the June 2026 refresh of Luma's Ray video model line, built with feedback from creatives across entertainment, advertising, and gaming. It powers four integrated workflows inside Luma's platform: text-to-video, image-to-video, video modification (video-to-video), and reframing. The headline advance over earlier Ray versions is frame-level direction: rather than a single start frame with an optional end frame and automatic in-betweening, Ray 3.2 lets you place up to 16 keyframes within one clip, giving production teams real authority over pacing, camera paths, and where each visual beat lands.
Ray 3.2 keeps HDR at the center of its output. It generates native 16-bit color and exports EXR files in the ACES2065-1 (AP0) color space, so footage drops straight into professional grading and compositing pipelines without a low-dynamic-range bottleneck. Rounding out the release are camera motion transfer for reusing cinematic moves across scenes, expressive performance tracking for up to eight faces at once, and an enhanced reframe tool for reshaping aspect ratio and background after the take. Ray 3.2 also marks the first time Luma has exposed this full control surface through a developer API.
Up to 16 keyframes can be placed within a single clip, letting you choreograph exact pacing, camera paths, and motion instead of relying on automatic in-betweening between a start and end frame.
The model generates 16-bit color natively and exports EXR in the ACES2065-1 (AP0) color space, making output ready for professional grading and VFX compositing workflows.
Cinematic camera moves can be captured and reapplied across different scenes, worlds, and styles, keeping a consistent directorial language between shots.
Performance tracking transfers full facial expression state, lip sync, and dialogue timing for up to eight faces at once, while preserving skeletal posture and gestures.
Shots can be reshaped after the take, adjusting aspect ratio and replacing background so a single generation adapts to multiple delivery formats up to roughly 12-16 seconds.
Text-to-video, image-to-video, video modification, and reframe share one model, so ideas can move from prompt to image-driven shot to edited output without switching engines.
Ray 3.2 is the first Ray release to expose its complete control surface through a developer API, designed to integrate into existing enterprise tools and pipelines.
| Modes | Text-to-video, image-to-video, video-to-video, reframe |
|---|---|
| Resolutions | 360p, 540p, 720p (default), 1080p |
| Text/Image-to-video length | 5 or 10 seconds |
| Modify Video length | Up to 20s (24fps=20s, 30fps=15s, 60fps=7s) |
| Reframe length | Approx. 12-16 seconds |
| Color depth | Native 16-bit HDR |
|---|---|
| Export format | EXR |
| Color space | ACES2065-1 (AP0) |
| Keyframes per clip | Up to 16 |
|---|---|
| Face tracking | Up to 8 faces simultaneously |
| Camera | Camera motion transfer |
| API | Full control surface available |
| Vendor platform | Luma (Dream Machine / Luma platform) |
|---|---|
| Released | June 9, 2026 |
| SharkFoto | Coming soon |
Frame-level keyframe control lets agencies land product reveals, logo beats, and camera moves on exact frames, then reframe one master into multiple aspect ratios for different placements.
Camera motion transfer and multi-keyframe pacing help directors block shots and build previsualization sequences with cinematic timing before committing to a full production.
Native 16-bit HDR with EXR export in ACES2065-1 drops generated footage straight into professional grading and compositing pipelines alongside live-action plates.
Expressive tracking for up to eight faces preserves lip sync and dialogue timing, supporting multi-character dialogue shots and performance-driven storytelling.
Consistent camera language across scenes and worlds, plus long native 1080p clips, suit in-engine-style cinematics, teasers, and trailer sequences.
| Feature | Ray3 | Ray 3.2NEW |
|---|---|---|
| Single start frame (optional end frame) with automatic in-betweening | Up to 16 keyframes per clip for frame-level direction | |
| Basic performance capture | Expressive tracking for up to 8 faces with lip sync and posture | |
| Shorter native 1080p clips | Up to 20 seconds (Modify Video, frame-rate dependent) | |
| Limited exposure | Full control surface via developer API | |
| HDR generation with EXR export | HDR generation with 16-bit EXR / ACES2065-1 export |
The 20-second maximum applies to Modify Video and depends on frame rate (24fps yields 20s, 30fps 15s, 60fps 7s). Text- and image-to-video clips are capped at 5 or 10 seconds.
Native output tops out at 1080p across the 360p/540p/720p/1080p tiers, so 4K delivery requires upscaling outside the model.
The 16-bit EXR / ACES2065-1 output is aimed at professional grading and compositing; teams without an HDR-aware pipeline may not realize its full benefit.
The reframe tool operates on roughly 12-16 second inputs, so very long source clips fall outside its supported window.
Frame-level keyframe control and native HDR video are on the way. Until Ray 3.2 lands, explore SharkFoto's available AI image and video tools, and check back here for launch updates.
Try Luma Ray 3.2 now