AI provider

Luma AI

Luma AI develops models for images and video. Founded in 2021, the company is based in Redwood City, California. Its cofounders are CEO Amit Jain and CTO Alex Yu. Its current model families include Ray for video and Uni for images.

Company & leadership

CEO
Amit Jain, cofounder.
Company
Luma AI, Inc.; privately held.
Funding milestone
$900 million Series C led by HUMAIN, reported in December 2025.

Founders

  • Amit Jain and Alex Yu.

From computer vision to creative models

Before Luma, Amit Jain worked on computer vision at Apple, including work associated with Vision Pro. Alex Yu left a PhD program to build the company after research in neural rendering and 3D reconstruction. These backgrounds help explain Luma’s early focus on capturing and generating three-dimensional scenes. Investor Andreessen Horowitz described Genie’s move from a November 2023 preview to general availability in January 2024 as an early product milestone.

International expansion and funding

Luma announced its first international office in London in December 2025. Former WPP executive Jason Day was appointed to lead its EMEA business. The announcement followed a $900 million Series C led by HUMAIN. This is a dated financing event, not a current estimate of the company’s value or cash balance.

In February 2026, Luma announced plans for a Riyadh office and work with HUMAIN Create on an Arabic-native foundation model. It also named Publicis Groupe as a partner. The announcement describes an expansion plan; it does not establish that an Arabic model is available through the public API.

How Ray and Uni fit into Luma’s history

Ray2 launched on 15 January 2025. Luma described a new multimodal architecture and increased training compute compared with Ray1. The release focused on motion, camera movement and coherent events within a clip.

Luma’s October 2026 training disclosure states that Ray3 began from third-party open weights and then received further training at Luma. It identifies Ray3.14 and Ray3.2 as descendants of Ray3, but does not name the original third-party model. The disclosure is useful evidence of model ancestry; it is not a license to download Luma’s finished models.

Uni uses a different approach for images. Luma describes an autoregressive transformer that processes text and image tokens in the same sequence. The Uni-1.1 announcement included image generation and natural-language editing, rather than treating the model as a video generator.

Current services and earlier model names

The current Agents API lists uni-1, uni-1-max and ray-3.2. Generation, editing and reframing are operations within that service. Separate guides below cover the older Ray releases, including their migration conditions. A model name in an older announcement does not mean that new accounts can still call it.

Luma now describes Photon as no longer being a separate product. Its current image offering uses the Uni family. The company’s older Genie and capture work remains part of its history, rather than evidence of a current image or video API endpoint.