Luma AI launches Ray3 video model
Luma AI has released Ray3, a video model it says reasons about and refines its own outputs and generates 16-bit HDR, targeting professional film and advertising pipelines through Dream Machine and Adobe Firefly.
Luma AI announced Ray3 on September 18, 2025, and made it available immediately in its Dream Machine platform. The Palo Alto lab calls it the first "reasoning" video model. It says the model can reason over visuals and concepts, evaluate its own outputs and refine them during generation.
The company also says Ray3 is the first model to output video in 10-, 12- and 16-bit High Dynamic Range in ACES2065-1 EXR format, the standard used in high-end film and advertising post-production. A new Draft Mode lets users explore ideas "up to 20x faster" before polishing chosen outputs to 4K HDR. CEO Amit Jain said Ray3 is "more than twice the size of Ray2" and claimed gains in fidelity, instruction following and temporal coherence. The announcement includes no benchmarks, architecture details or independent evaluations to back up these claims.
Adobe is the first partner to offer Ray3 outside Dream Machine. It is available globally in Firefly's Video module and Firefly Boards, and its outputs can be moved into Premiere Pro. Customers on paid Firefly or Creative Cloud Pro plans get unlimited Ray3 generations for their first 14 days.
Luma also named several agency launch partners:
- Dentsu Digital, the sole launch partner in Japan, which will build Ray3 into its ad production pipelines
- HUMAIN Create, which is bringing the model to creators and enterprises in the MENA region
- Monks, Galeria and Strawberry Frog
Luma did not disclose pricing for Dream Machine.