Spotlight: Ray 3.2 — AI video that keeps its HDR data intact for VFX
Most AI video generators hand you an MP4 and consider the job done. Ray 3.2 is built for the step after that: it generates clips in real 16-bit HDR (not a LUT slapped on top) and exports as EXR in the ACES2065-1 color space, so the footage can go straight into Nuke or DaVinci Resolve and survive an actual grading pass. That's a narrower promise than "make a cool video," and it's the right one to make.
The product itself is four workflows stitched into one browser tool: text-to-video, image-to-video, a video-edit mode that restyles existing footage while keeping the original motion and timing, and a reframing mode that reshapes a clip for different aspect ratios. Clips run up to 10 seconds for text/image generation and up to 20 seconds for edits, which tells you this is aimed at spots, social cuts, and concept previz rather than anything long-form.
The smart part isn't the model itself. Ray 3.2 is Luma AI's model, and it's already available through Adobe, Freepik, Fal.ai, Comfy, Runware, and a handful of other platforms. What this site is selling is a direct, credit-based front end to that model, with the specific pitch aimed at people who need frame-level control and pipeline-ready files rather than a quick clip to post. Up to 16 keyframes per clip means you can direct where a look shifts or a performance lands within a single generation instead of stitching together separate short clips and hoping the seams don't show. Credits don't expire and are only deducted on success, which is a small but real detail: you're not burned by failed generations, which is common friction in this category.
Who should try this: marketing and creative teams making campaign variants from existing footage, video editors who need to reframe one hero clip into six aspect ratios without reshooting, and small production teams who want AI-generated plates or previz that a colorist can actually grade instead of fighting flattened, crushed color. If your deliverable ends up in Resolve or Nuke, the EXR/HDR pipeline is the actual reason to pick this over a generic text-to-video tool.
Who should skip it: if you just want a fast, throwaway clip for a TikTok and don't care about color depth or VFX compatibility, you're paying for capability you won't use. There's no audio or sound generation mentioned anywhere in the feature list, so this is silent footage you'll still need to score and mix elsewhere. And at 10 to 20 seconds max, it's not a fit for anyone trying to generate longer narrative sequences in one pass.
The pricing is tiered credit packs, with the higher tiers advertising several times the credits of the entry tier plus priority processing, which suggests the entry plan is really a trial run before you commit to production volume. Worth testing on the cheapest tier first to see how the HDR/EXR round-trip actually behaves in your specific pipeline before buying up.
Where this probably goes next, and this is a guess: longer clip durations feel like the obvious ask given how tightly the current limits are set, and some kind of native audio or sound-design pass would close the biggest gap for anyone trying to ship a finished spot rather than a silent plate. The more interesting bet is whether "AI output that survives a real grading pipeline" becomes table stakes across this category, in which case Ray 3.2's early positioning here is worth watching rather than a one-off gimmick.
Try Ray 3.2: ray32.net
See the launch: Ray 3.2 on welaunch.sh