Luma's Ray2 launched to paid users only, no API yet

The shot

In January 2025, Luma AI published a dedicated product page for Ray2, the model succeeding Dream Machine, the launch already covered on this site. Archived at the time, the page states Ray2 exhibits advanced capabilities as a result of training on a new multi-modal architecture 'scaled to 10x compute of Ray1,' and that the model was built to produce 'fast coherent motion, ultra-realistic details, and logical event sequences.' The page states plainly: 'Ray2 is available today in Dream Machine, starting with paid subscribers,' with image-to-video, video-to-video and editing capabilities named as 'coming soon' rather than available at launch.

What the documents show

The '10x compute of Ray1' figure and the motion-quality language are Luma's own stated claims about its training process, not an outcome an outside evaluator measured on this page. Luma's news index, archived some months later, corroborates Ray2 as an ongoing product line rather than a single release: it logs a 31 March 2025 update, 'Introducing Camera Motion Concepts,' described as 'now available for Ray2,' showing the model continued receiving named feature additions after its initial rollout rather than shipping complete on day one.

The workflow

At launch, Ray2 reached only paid Dream Machine subscribers through text-to-video generation; a production wanting image-to-video, video-to-video or editing through Ray2 had to wait for those modes to ship separately, and the page states API access specifically was still pending ('Ray2 will be available in the Luma API soon'). A pipeline built around Ray2 at this stage therefore needed a subscription-based interactive workflow rather than a scriptable, automated one, since the programmatic access path had not yet opened.

What the tool does not change

Luma's own description of 'logical event sequences' is a stated training goal, not a guarantee that a specific requested sequence of actions will render coherently on a given attempt; the site's own examples remain selected demonstrations rather than a documented success rate. Nothing on the page describes an editorial or continuity-checking function, so a team still needed to review generated shots against the surrounding footage's motion and lighting before treating a Ray2 clip as production-ready.

• Which specific modes, text-to-video, image-to-video, editing, were actually live on the date being cited, rather than announced as coming soon.
• Is a stated compute multiple, such as '10x,' a claim about training cost or about output quality, and does the source distinguish the two.
• Has an access path such as API availability actually shipped, or does the source still describe it as pending.

Luma's own product page and news index date Ray2 to a paid, subscriber-only rollout inside Dream Machine in January 2025, with programmatic access and further modes added only in the months that followed.