Background
Jain founded Luma with Alex Yu, building from neural radiance fields and 3D capture (Genie) into video generation (Dream Machine, the Ray line) and now multimodal world models. Luma's Series B was led by a16z; the Series C in November 2025 brought in HUMAIN and a pivot toward multimodal foundation models at much larger compute.
He talks publicly about data curation as the central lever, rejecting the throw-garbage-at-it approach, and about an internalisation strategy — capabilities first handled by external scaffolding get absorbed into the next model generation.
What they run now
Luma's whole business: the Ray video line (Ray3, Ray3 Modify, Ray3.14), the Uni-1 image model, Dream Machine, Dream Lab LA as a creator studio, and the international expansion under Jason Day in London. The company positions itself around professional video — its evaluation reports are aimed at filmmakers and studios — while the research is moving toward world simulation.
What they buy
Inference. Jain is the economic buyer: curated licensed video, expert evaluation panels, and — given Dream Lab LA and the Forward Deployed Creative team — professional creative input that shapes the product. His internalisation view is a caution: he prefers building capability into the model over paying external systems, which means a data vendor has to be framed as an input to the model, not as an ongoing outsourced service. Note the HUMAIN relationship raises the salience of Arabic-language and regional creative data, an angle no one in this batch is publicly serving.
Career
- 2021 – presentCo-founder & CEO, Luma AI — Palo Alto / Redwood City; offices in London and Riyadh
On the record
Says the job is to show the model what you want it to learn and show it good examples, not to feed it everything. source
Argues anything that can be done inside the model will beat an external system doing it. source
Says the stronger the model's foundational understanding, the better it handles out-of-distribution situations. source