Luma's 2026 timeline: Ray3.14, Ray3.2, Scenes and Variants
Luma shipped Ray3.14 in January, Ray3.2 in June, Scenes in August and Variants on Oct 1, 2026. What each added, and why to pin model ids.

Luma released four notable things in 2026: Ray3.14 in January, Ray3.2 in June, Scenes in August, and Variants on Oct 1. The Luma news page describes each in a line, and the dates show a release every two to five months. That pace is why you should pin model ids in code and log which one produced each clip.
What shipped, in order
The table lists what the news page states for each release, nothing more.
| Release | Date | What the page says |
|---|---|---|
| Ray3.14 | Jan 2026 | Native 1080p, 4x faster, 3x cheaper |
| Ray3.2 | Jun 2026 | Frame-by-frame direction |
| Scenes | Aug 2026 | Storyboard review before render |
| Variants | Oct 1, 2026 | Resize and translate static ads into every format while keeping logos, headlines, and CTAs |
Which ones change your workflow
Three of the four are model or workflow releases, and one, Variants, is aimed at static ads. Scenes is a process change: it adds an approval step before the expensive render. That pattern is worth copying whichever model you use.
Pin the id
When a vendor ships this often, the same prompt can produce a different look in the next version. Pin the model id in your request, store it with the job id, and upgrade on purpose. On Sume, the video docs say a pinned catalog id runs that model, while sume/auto echoes sume/auto and never discloses which family served the request, so do not use Auto where you need to reproduce a look.
{
"model": "seedance-2.5",
"prompt": "Product on a desk, slow push-in",
"duration": 6,
"resolution": "720p"
}A change log for your own prompts
Keep a small change log of the prompts that matter and the model id that produced the approved clip. When a vendor ships a new version, rerun the log's prompts and compare. This is how you find out whether an upgrade helps your content before you switch a production path.
Treat the vendor's own claims, such as faster or cheaper, as hypotheses to test. The Ray3.14 line on the news page says 4x faster and 3x cheaper, and the only way to know what that means for your clips is to time and price a pilot.
- Store prompt, model id, and date for approved clips.
- Rerun on every new version.
- Compare cost, time, and look.
- Switch deliberately, not by default.
Where to look next
Read the live catalog for what Sume lists today. The video docs describe how to compare ids, ranges, and prices across models, and the stored posts on Scenes and Variants cover those two features.
Sources
Related posts
More in Models
- Lyria 3.5 blocks artist-voice prompts: how to write briefs that pass
Google's Lyria 3.5 docs note that prompts asking for specific artist voices are blocked. Describe the sound instead, then run it through the Sume Music Router.
- MiniMax H3 limits: 9 images, 3 videos, 3 audio, file caps
MiniMax's H3 guide caps prompts at 7,000 characters and references at 9 images, 3 videos and 3 audio files. Cheat sheet with the Sume limits beside it.
- Nano Banana reference image limits: Lite, 2 and Pro compared
Nano Banana 2 Lite takes up to 14 object images, Nano Banana 2 takes 10 object, 4 character and 3 style, Pro takes 6 object and 5 character.
- PixVerse V6 native audio and camera work: what to check in an API
PixVerse's blog lists V6 with camera work and native audio, plus R2 and a $439M Series C total. How to test those claims against any video API's catalog.
Written by Sume