Luma Ray 3.2 Prompt Generator

Verdict

Feature snapshot

Ray 2 → Ray 3.2: what changed (HDR is the headline)

Ray 2 was already a big step over Luma's older models, strong prompt understanding, lots of camera motion, a genuinely cinematic feel, but it was unstable: body parts deformed, walk cycles glided, and at 720p with 5–10s clips the picture often started to warp and shift around the 6–7 second mark. Ray 3.2's headline upgrade fixes the look rather than the physics: native HDR. In HDR the bright parts actually pop and the dark areas read as deep blacks instead of the flat, TV-like gray that SDR gave you, which is what makes a Ray 3.2 clip look 'straight out of a movie.'

The practical win is grading headroom. Reviewers shoot 1080p HDR specifically so they have more room to color grade in post-production, and the side-by-side SDR-vs-HDR shots show the whole aesthetic filling out with depth. Ray 3.2 also tightened prompt adherence over Ray 2 and handles longer clips and extensions more gracefully, though, as below, 'better' is not 'fixed.'

How Luma compares to other AI video models

Where Luma Ray 3.2 sits against the rest of the field on value and output quality, and how it scores capability by capability. Hover or tap any model for the detail.

Keyframes for transitions, loops, and product shots

Luma's most distinctive control is keyframes: you set a starting clip and an ending clip, and the model fills the motion between them. Ray 3.2 extends this with multi-keyframe support so you can pin several story beats across a shot, reviewers describe it as 'directing, not just guessing,' with story beats locked so a character does exactly what you intended instead of morphing.

This is exactly why the model leans toward product videography and clean transitions. In the product test, uploading four images as exact keyframes made Ray hit every visual beat in order. The catch: it hit the keyframes but flattened the in-between into a simple fade and skipped a prompted marble-shatter, so keyframes control where you land, but the transition you describe in prose still needs to be simple and explicit, not a complex effect.

Where Luma shines: cinematic stills, close-ups, materials

Luma is at its best on a single, beautiful, contained shot. Close-ups of faces and upper torsos hold up well, Ray 2 nailed an extreme sad expression with real raw emotion, and Ray 3.2 keeps faces consistent even while a character speaks. Materials and light are its showcase: flowing fabric, a record spinning under a needle with dust catching the light, subtle reflections on glossy surfaces. The 'fabric woman' clip ripples with no morphing and no warped limbs.

It also reads prompts adventurously, which is a creative strength: ask for origami birds and Luma actually folds them out of paper where a more literal model just renders ordinary birds. Lean into this, give Luma an evocative, art-directed single shot and let its lighting do the work.

The limits: no audio, deformation, and longer clips

The biggest gap in 2026 is audio: Ray 3.2 generates none. Reviewers call it 'surprising' that a brand-new model ships silent when every competitor has sound, and in the head-to-head the silence directly hurt Luma, Seedance's clips felt much stronger largely because of their sound design. If your project needs dialogue or SFX, you'll be adding it separately.

The other limits are motion and logic. Complex full-body movement still triggers deformation and the infamous six fingers; an airport emotion test froze every background character. Ray 3.2 also has no video reference, general image reference, or audio reference inputs and no native multi-shot generation, and longer clips still tend to drift, Ray 2's 10s clips warped around 6–7s. The takeaway: keep each Luma shot short, contained, and physics-light.

Luma vs Seedance

In the direct test, the split was clean. On pure visual quality, lighting, color, naturalness, Ray 3.2 looked better, even more natural than Seedance 2.0 in places, and it won the text-to-video look outright. But Seedance followed the prompt more faithfully (it delivered the requested full trot and tracking camera where Luma just wandered), it carries native audio, and it supports the multi-shot and reference inputs that became standard after Seedance 2.0.

So the rule of thumb: reach for Luma when you want the most cinematic single shot, an HDR product or transition piece, or a strong close-up, and you can live without sound. Reach for Seedance when prompt-faithful action, audio, or multi-shot control matter more than the prettiest frame. Use this generator to write the most Luma-flattering version of your idea regardless of which way you lean.

Other model prompt generators