Frontier AI lab behind Dream Machine — multimodal AGI through video, image, and 3D generation
HQ: San Francisco / Palo Alto, CA | Founded: 2021 | Employees: ~250 (2026, est.) | Stage: Late Stage VC (Series C) | Website: https://lumalabs.ai
Luma AI was founded in 2021 in the San Francisco Bay Area by Amit Jain (ex-Apple camera and computer-vision engineer, later of Product Hunt and Circle Medical) and Alex Yu (UC Berkeley PhD who came out of NeRF research). The company is self-described as an AGI lab rather than a video-tool company — its stated ambition is systems trained on a quadrillion tokens of multimodal information (video, image, audio, language).
Luma began in neural rendering and 3D capture (the Luma app captured real scenes via NeRF/Gaussian Splatting from a phone), then shipped Genie, a text-to-3D generator (research preview November 2023, Genie 1.0 with its Series B in January 2024). Its breakout came on June 12, 2024 with Dream Machine: a free AI video generator that went instantly viral, drove multi-hour wait queues, and within six months pushed the company past 25M users. The Dream Machine platform (web + iOS) added Photon, a state-of-the-art Universal-Transformer image model (November 2024), the Ray2 video model (announced at AWS re:Invent, launched January 2025 at 10x Ray1 compute), and subscriptions. In September 2025 Luma launched Ray3, billed as the world's first reasoning video model — it plans, storyboards, and self-evaluates before generating, produces high-fidelity 16-bit HDR in professional ACES2065-1 EXR, and launched inside Adobe Firefly as the first non-Dream-Machine partner — followed by Ray3.14 (January 2026: 4x faster, 3x cheaper) and Ray3.2 plus the first full Ray-control API (June 2026).
Commercially, Luma is anchored by strategic infrastructure: a December 2024 partnership brought Ray2 to Amazon Bedrock ('the first and only cloud provider to offer fully managed Luma models') with compute and custom-chips backing; the November 2025 $900M Series C led by HUMAIN (Saudi Arabia's PIF AI company) linked Luma to Project Halo, a 2-gigawatt AI supercluster in Saudi Arabia, with Luma as anchor customer locking in 100 MW. Additional partnerships include Publicis (MENA preferred AI partner), Dentsu Digital (Japan), Serviceplan, and a Hollywood production joint venture (Innovative Dreams, with Wonder Project) whose first project, 'The Old Stories: Moses' starring Ben Kingsley, heads to Amazon Prime Video. Luma grew from ~130 to ~250 people through 2026, opened Dream Lab LA and a London EMEA hub, and maintains an 'Open Physical AI Lab'. Its revenue is early (FY2024 revenue reported ~$8M; est. ~$21M annualized by mid-2025) against heavy compute spend, funded by the >$1B raised across seed through Series C.
Luma is the rare consumer-AI story that converted virality into one of the biggest funding rounds in generative media: 30M+ users, Dream Machine's breakout summer of 2024, and a $900M Series C at a >$4B valuation (Nov 2025) with Saudi PIF's HUMAIN and 2-gigawatt supercluster access behind it. The bull case is that Luma pairs best-in-class video quality with an actual path to 'world models' — reasoning models that generate physical, plausible sequences — which is exactly the substrate AGI needs, plus enterprise distribution through Amazon Bedrock and Adobe. The bear case is brutal competition (Google Veo, Runway, Kling, Midjourney), massive compute burn that depends on sovereign superclusters, thin near-term revenue (~$8M in 2024), and regulatory pressure (California's SB 942 transparency regime) on top of unresolved copyright questions.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki