Drop in a clip, set up to 16 keyframes, and direct motion, lighting, and character the way you would on a real set. Built on Luma's Ray 3.2 model, designed with studios, agencies, and game teams.
What is Ray 3.2 AI Video Generator
A Ray 3.2 AI Video Generator is a tool, built on Luma's Ray 3.2 video model, that transforms an existing video clip frame by frame while keeping its original length, structure, and motion.
Ray 3.2 is a video-to-video model from Luma Labs, released June 9, 2026, and designed in collaboration with creatives from the entertainment, advertising, and gaming industries (source: Introducing Ray 3.2). Unlike a text-to-video tool that invents a scene from scratch, the Ray 3.2 AI Video Generator takes a clip you already have and lets you restyle it, relight it, or recast it - without losing the original timing.
Three things separate Ray 3.2 from earlier AI video models:
In short: the Ray 3.2 AI Video Generator is the production-grade V2V workhorse for transforming existing video while keeping the structure of the source (paraphrased from Luma Ray 3.2 core concepts).
Key Features
Every feature in the Ray 3.2 AI Video Generator was built around one idea: keep the structure of the original clip, change everything else. Here is what you actually get.
Place up to 16 keyframes inside a single clip and lock pacing, blocking, or character expression on each one. Ray 3.2's underlying model accepts up to 64 source-frame indexes for fine art direction (source: Luma Ray 3.2 core concepts).
Restyle, recolor, or repurpose any clip up to 20 seconds long at 1080p. Source duration and motion are preserved exactly, so a 12-second take comes back as a 12-second take (source: Luma Ray 3.2 product page).
HDR generation is native, not upscaled. Export 16-bit frames in ACES2065-1 EXR for direct color grading in Nuke, Resolve, Flame, or Baselight (source: Luma Ray 3.2 prompting outputs and controls).
Two 1-9 sliders. Motion controls how closely the result follows the movement in your source clip. Structure controls how closely it follows the spatial layout (source: Luma Ray 3.2 prompting outputs and controls).
Lock facial identity, body proportions, or full skeletal pose. Use Blocking for stylized non-human transformations where you want flexibility, not a rigid skeleton (source: Luma Ray 3.2 prompting outputs and controls).
Reshape a shot after it has been generated. Adapt aspect ratios for vertical, square, or letterboxed platforms, extend the frame, or replace a background - without rerunning the take (source: Introducing Ray 3.2).
Maintain skeletal posture and gestures while tracking the full expressive state for up to eight faces simultaneously, frame by frame (source: Introducing Ray 3.2).
Capability gallery
Eight built-in transformations cover the work most teams hire post for. Each tile below is one capability of the Ray 3.2 AI Video Generator, with a sample clip and what it does to your source take.
An 8-tile gallery. Each tile = sample clip + H3 + 1-2 sentence description + suggested image alt. Sources: all eight capabilities are described on Luma Ray 3.2 product page.
Carry movement and dynamics from one clip onto a new character, object, or material. The Ray 3.2 AI Video Generator keeps the rhythm of the original take while everything visible can change.
alt: Ray 3.2 AI Video Generator motion transfer example
Reuse cinematic camera moves across new scenes, worlds, and styles. The dolly, push-in, or orbit from one shot becomes a repeatable move on the next, applied through the Ray 3.2 AI Video Generator.
alt: Ray 3.2 AI Video Generator camera motion transfer example
Turn one person into an entirely new character while preserving their performance. The Ray 3.2 AI Video Generator keeps facial expression and body motion locked, then swaps look, costume, or species.
alt: Ray 3.2 AI Video Generator character transformation example
Add cinematic effects and transformations while keeping scene consistency. The Ray 3.2 AI Video Generator works on the source clip directly, so the effect lines up with the take you already cut.
alt: Ray 3.2 AI Video Generator visual effects example
Move a scene to any setting, season, or world without rebuilding the shot. The Ray 3.2 AI Video Generator preserves the subject and the camera while the world around them changes.
alt: Ray 3.2 AI Video Generator environment change example
Re-light a finished take to match a new environment, time of day, or creative direction. The Ray 3.2 AI Video Generator keeps the source motion and structure; only the lighting shifts.
alt: Ray 3.2 AI Video Generator relighting example
Adapt products across campaigns with no reshoot. The Ray 3.2 AI Video Generator places a new product into the same scene, lit and framed the way the original was shot.
alt: Ray 3.2 AI Video Generator product swap example
Generate market-specific campaign variations from one source take. The Ray 3.2 AI Video Generator handles localization - product, environment, or talent - so one shoot covers every region.
alt: Ray 3.2 AI Video Generator global campaigns example
All eight transformations are documented on the official Ray 3.2 page (source: Luma Ray 3.2 product page).
How It Works
Most clips finish in under five minutes. You bring the source take and a written description of what you want; the Ray 3.2 AI Video Generator handles the rest.
Drop in any video up to 20 seconds. Ray 3.2 is a video-to-video model, so it works from a clip you already have, not from a blank prompt (source: Luma Ray 3.2 core concepts).
Describe the end result, not the change. "A purple sky over the mountain landscape" works; "change the sky to purple" does not. Avoid negatives like "no," "not," or "without" (source: Luma Ray 3.2 prompting outputs and controls).
Open the keyframe editor and pin up to 16 reference frames. Each keyframe can carry its own instruction and character reference so the Ray 3.2 AI Video Generator hits exact beats - entrance, midpoint, end (source: Luma Ray 3.2 core concepts).
Use the 1-9 sliders. Higher values keep the result closer to the source clip; lower values give the model more freedom for stylized transformations (source: Luma Ray 3.2 prompting outputs and controls).
Choose 360p draft for fast iteration or 1080p HDR + 16-bit EXR for the final cut. The same Ray 3.2 AI Video Generator handles draft and finish without re-rendering the take (source: Luma Ray 3.2 product page).
At the end you get a finished clip in your chosen resolution and format, with the original duration and motion intact and ready to drop into the next step of post.
Use Cases
Ray 3.2 was built with studios, agencies, and game teams. The Ray 3.2 AI Video Generator fits anywhere a source clip needs to be reshaped instead of rebuilt.
Source: Introducing Ray 3.2.
Take a take that almost worked and finish it. Relight a scene, swap an environment, or carry a complex actor performance into a new look using Performance Tracking. Output goes straight to EXR for color grading (source: Luma Ray 3.2 product page). The Ray 3.2 AI Video Generator slots into existing post pipelines instead of replacing them.
Spin one shot into a global campaign. Use Reframe to adapt a single hero clip to vertical, square, and letterbox in one pass, and Product Swap to localize the on-screen product per market (source: Luma Ray 3.2 product page). One source take, dozens of platform variants.
Explore looks before committing to render. Use the Ray 3.2 AI Video Generator for VFX exploration, look development, and character transformation on existing animatics. Source duration is preserved so the result lines up with your edit (source: Luma Ray 3.2 core concepts).
Re-shoot is expensive. The Ray 3.2 AI Video Generator handles relighting, environment change, and product swap on existing product footage so a single source take can serve a full season's worth of variations (source: Luma Ray 3.2 product page).
Compare
Where the Ray 3.2 AI Video Generator fits among Luma's own models and the other 2026 leaders. Comparison axes: input, max length, HDR, keyframes, API.
| Ray 3.2 AI Video Generator | Ray 3 | Ray 3.14 | Other 2026 leaders | |
|---|---|---|---|---|
| Primary input | Video-to-video | Text + image + video | Text + image + video | Mostly text + image |
| Max clip length | 20 seconds at 1080p | Shorter | Up to 18 seconds | 10-20 seconds typical |
| HDR generation | Native 16-bit | Native 10/12/16-bit | Native HDR | Limited or absent |
| EXR export | 16-bit ACES2065-1 | EXR available | EXR available | Not standard |
| Keyframes per clip | Up to 16 (product) / 64 (model) | Start + end | Start + end (Modify) | None to limited |
| Built-in transformations | 8 (motion transfer, relighting, product swap, environment change, etc.) | Reasoning-driven, fewer V2V tiles | Modify start/end only | Mostly text-to-video |
| Best for | Restyle, rescue, localize existing clips | Reasoning-driven first drafts | Faster 1080p iterations | Generative scenes from scratch |
Sources: Luma Ray 3.2 product page, Introducing Ray 3.2, Luma Ray, Ray 3.14 reference, Luma video models guide.
When the Ray 3.2 AI Video Generator is the right call: you have a source clip and you need to change it. When something else is the right call: you have no clip and need to generate one from a prompt - Ray 3 or a text-to-video model fits better there.
Why choose
Plenty of tools generate video. Few of them direct it. The Ray 3.2 AI Video Generator is built for the moment after the first prompt - when you need a specific result.
Most AI video tools give you a roll of the dice on each prompt. The Ray 3.2 AI Video Generator hands the dice back. Up to 16 keyframes per clip, two adherence sliders, and four character locks mean the next take looks like the one you asked for (source: Luma Ray 3.2 prompting outputs and controls).
Native 1080p, 16-bit HDR, and ACES2065-1 EXR export mean no conversion step before the color suite. The Ray 3.2 AI Video Generator was designed to be one stop in a pipeline, not the whole pipeline (source: Luma Ray 3.2 prompting outputs and controls).
A 12-second source clip comes back as a 12 second result. Edits and timing already locked in your cut do not need to be reworked (source: Luma Ray 3.2 core concepts).
Use 360p draft for cheap iteration and 1080p HDR + EXR for finish. Same Ray 3.2 AI Video Generator, same controls, no re-prompting the work (source: Luma Ray 3.2 product page).
Source: Luma Ray 3.2 product page.
Trusted by the studios, agencies, and platforms shipping cinema.
Trust line (verbatim from Luma). Source: Luma Ray 3.2 product page.
Designed in collaboration with creatives from the entertainment, advertising, and gaming industries.
Positioning quote (paraphrased from Luma). Source: Introducing Ray 3.2.
FAQ
The most common questions about the Ray 3.2 AI Video Generator, answered from Luma's official model and product documentation.
The Ray 3.2 AI Video Generator is a video-to-video tool built on Luma's Ray 3.2 model. You upload a source clip, write a target-state prompt, and the model returns a new version of the clip with the original duration, motion, and structure preserved (source: Luma Ray 3.2 core concepts).
Ray 3.2 was introduced by Luma Labs on June 9, 2026, as an update to Ray 3 (source: Introducing Ray 3.2).
No. Ray 3.2 is a video-to-video model. It needs a source clip to work from. For text-to-video, Luma's Ray 3 is the right model (source: Luma Ray 3.2 core concepts).
Up to 20 seconds at 1080p in Modify Video V2. Source duration is preserved exactly (source: Luma Ray 3.2 product page).
The Ray 3.2 AI Video Generator product workflow exposes up to 16 keyframes per clip. The underlying model accepts up to 64 source-frame indexes for fine art direction (source: Luma Ray 3.2 core concepts).
SDR, HDR, and HDR + 16-bit EXR (ACES2065-1). Resolutions go from 360p draft to 1080p (source: Luma Ray 3.2 product page).
Eight built-in transformations: Motion Transfer, Camera Motion Transfer, Character Transformation, Visual Effects, Environment Change, Relighting, Product Swap, and Global Campaigns (source: Luma Ray 3.2 product page). All of them preserve the source clip's duration, motion, and structure.
Pricing on Luma is credit-based, with multipliers for resolution, length, and HDR/EXR output (source: Luma Ray 3.2 product page). On the Ray 3.2 AI Video Generator above, the free trial covers a starter pack of credits - enough to finish a first clip.
Bring one clip. The Ray 3.2 AI Video Generator handles the rest - keyframes, HDR, EXR, all in one place.
Free starter credits. No card required to try Ray 3.2.
Direct any frame. Finish every cut. Source: Luma Ray 3.2 product page.
Social Proof
Trusted on the Ray 3.2 model that ships cinema
The Ray 3.2 AI Video Generator is built on Luma's Ray 3.2 - the model Luma describes as "trusted by the studios, agencies, and platforms shipping cinema".