Luma's director-grade model: pin up to 16 keyframes in a single clip, keep performances intact across 8 faces, and export HDR that drops into a real grading pipeline. Type a prompt below and try it.
Developer
Luma Labs
Type
Text / image / keyframe to video
Keyframes
Up to 16 per clip
Duration
5–10 s core
Output
Up to 1080p · native HDR
Post pipeline
16-bit EXR · ACES2065-1
Tier here
Paid credits
Our one-line verdict: the post-production pick. When the clip must land exact beats and then survive a real color grade, nothing else here speaks EXR.
Ray 3.2 shipped in June with a pitch aimed past prompters at directors: place up to 16 keyframes inside a single clip and choreograph exactly where each beat, camera move and look-change lands — the largest anchor count we host, ahead of FLUX 3's ten. The second half of its case is the part no rival touches: native HDR generation with 16-bit EXR export in ACES color, meaning the output drops into a professional grading pipeline instead of arriving as a baked, finished-looking file. It was built with studios and agencies in the room, and it shows.
Judged from our own renders and the published record, revised as the facts move — that's a feature of this page, not a disclaimer.
• 16-keyframe direction.
The most anchors of anything we host. Storyboard-first teams pin the beats; the model animates between them with real authority over pacing and camera paths.
• HDR + EXR for real pipelines.
Native high-dynamic-range generation and 16-bit EXR export in ACES2065-1 — the only output here a colorist can grade rather than merely accept.
• Performance tracking.
Posture, gestures and full facial expressiveness carried across takes, for up to 8 faces at once. Ensemble scenes keep their acting.
• The Edit sibling.
Ray 3.2 Edit re-imagines footage you already have with the same frame-level control — revision as direction, not another spin of the slot machine.
• Silent output.
Like Runway, Ray 3.2 ships without audio in an audio-native year. Budget the sound pass — or pick FLUX 3 when synced audio matters more than anchors.
• Grounded beats imaginative.
Independent reviewers found it strongest on clean, physically plausible direction — and weaker when asked to think like an animator or bend physics. Stylized fantasy briefs fit FLUX 3 better.
• Short core durations.
5–10 seconds per generation on the surfaces we run (some platforms stretch longer). Long-take briefs belong to the 30-second models.
• Pro features, pro invoice.
HDR and EXR renders price above standard output, and sixteen keyframes invite over-direction. Start with three anchors; earn your way to sixteen.
Copy into the generator above, swap the brackets, render.
Pin 5 keyframes: wind-up → leap peak → landing → held pose → exhale → "A dancer alone in a rehearsal room, one shaft of window light. The camera drifts a slow half-circle across all five beats."
Keyframes carry the timing so the prompt only sets place, light and camera. Odd-numbered anchor counts (3, 5, 7) read as rhythm; sixteen is a ceiling, not a target. The demo beside the generator is this prompt.
"Macro: a perfume bottle on black glass, one hard rim light. The camera orbits 90° as a second, warmer light fades in. HDR on, EXR export on."
Shoot it flat-intent like a real set: controlled lights, no baked style. The EXR lands in ACES so your colorist grades the look instead of fighting one — the workflow no other model here can enter.
Upload: a reference clip of an actor's delivery
"Same performance, new setting: the monologue now lands in a rain-streaked parking garage at night. Keep every gesture, glance and pause exactly as tracked."
Performance tracking is the feature — 'exactly as tracked' invokes it. Re-staging beats re-generating: the acting survives, only the world changes.
The structural-control tier: three models for teams that storyboard first. Same prompts across all three, judged by what we'd ship. All three selectable in the generator above.
16-keyframe frame-level direction, HDR/EXR output, performance tracking, the Edit variant, and the first full Ray API. Built with entertainment, advertising and gaming creatives. The version in the generator above.
the generation that made HDR the family's identity: the first video models producing true high-bit-depth output in professional EXR standards.
Luma's consumer era, when the pitch was accessibility. The 3.x line is the same lab deciding its real customer is the production pipeline.
Luma Labs' video model, announced June 9, 2026, built for directed generation: up to 16 keyframes placed inside one clip for frame-level control of beats and camera paths, performance tracking across up to 8 faces, and native HDR output with 16-bit EXR export in ACES2065-1 for professional grading pipelines.