Luma AI

Luma AI
Information
Type	Private
Industry	Artificial intelligence, 3D capture, generative video
Founded	2021
Founder	Amit Jain, Alex Yu, Alberto Taiuti
Headquarters	Palo Alto, California, United States
Notable Personnel	Amit Jain (CEO)
Products	Luma 3D Capture app, Luma Unreal Engine plugin, Genie, Dream Machine, Ray video models
Website	https://lumalabs.ai

Luma AI (Luma AI, Inc.) is an American artificial intelligence company headquartered in Palo Alto, California. It started in 2021 as a 3D capture company built on Neural Radiance Fields (NeRF), letting people turn ordinary phone footage into photorealistic, navigable 3D scenes, and later moved into generative 3D and generative video.

For VR and AR, Luma's early work is the relevant part. Its capture app, web app and API produced NeRF and Gaussian splatting scenes that could be exported to game engines, and its free Unreal Engine plugin was presented as a way to bring volumetric captures into games, virtual production and virtual reality experiences. Since the launch of the Dream Machine video generator in June 2024, the company has concentrated on video models (the Ray series) and on multimodal "world models". In November 2025 it raised a 900 million US dollar Series C led by the Saudi AI company HUMAIN, which CNBC reported valued the company at more than 4 billion US dollars.

Reviewed 6 October 2026. Founders, funding rounds, offices, App Store listing, luma-web VR note, and Genie, Dream Machine and Ray release claims checked against the cited sources. About review dates.

History

Founding and NeRF capture (2021-2022)

Luma was founded in 2021. When TechCrunch reported its seed round in October 2021, it named co-founders Amit Jain as CEO and Alberto Taiuti as CTO and described the founders as engineers who had left Apple's AR and computer vision group. Andreessen Horowitz later described Jain as having worked on the Apple Vision Pro and on Apple's computer vision team. The third co-founder, Alex Yu, came from UC Berkeley, where he did neural rendering research with Angjoo Kanazawa. His personal site describes him as Luma's former co-founder and CTO and dates his time at the company to 2022-2024. Yu was the lead author of PlenOctrees, an ICCV 2021 paper that rendered NeRF scenes in real time, at more than 150 frames per second for 800x800 images, and a co-author of Plenoxels (CVPR 2022), which replaced the neural network in a radiance field with a sparse voxel grid and optimized about two orders of magnitude faster than NeRF.

The 4.3 million US dollar seed round was led by Matrix Partners, with South Park Commons, Amplify Partners, Context Ventures, Andreas Klinger and angel investors taking part. At that stage Luma pitched its neural rendering as a cheaper way to make photorealistic 3D product models for online retail: in its example, 20 of 30 photos of a car trained the network and the remaining 10 were used to check its accuracy. TechCrunch noted that such a model could also be viewed in a VR headset, and the company planned a YouTube-like embeddable 3D player for retailers.

Capture app, Unreal plugin and Gaussian splats (2023)

Luma closed a 20 million US dollar Series A in March 2023. Amplify Partners led it, with General Catalyst and NVIDIA's venture arm joining and Matrix Partners and South Park Commons continuing.

By April 2023 Luma offered an iOS app for capturing with an iPhone and a web app that accepted uploaded video or zipped image sets, both free in beta. Web app users could download object captures in glTF, OBJ or USDZ format, scene captures in PLY format, and Luma Field files for use in Unreal Engine. On 4 April 2023 it released a free alpha plugin for Unreal Engine 5 that rendered these volumetric NeRF captures inside the engine in real time. The plugin created Blueprints with baked or dynamic lighting, could crop objects automatically and could add a sky environment. The first release ran only on Windows and did not generate colliders. Version 0.3 followed in July 2023 with rotated crop boxes and quality presets from Minimal to Extreme.

In early October 2023 Luma introduced Interactive Scenes, its own version of Gaussian splatting (a rendering technique from a research paper by Inria). Radiance Fields, which tested it before launch, described it as a proprietary iteration of the original Inria method that combined real-time splat rendering with Luma's cloud-based rendering. The scenes ran in browsers and on phones, and Luma said they streamed as files of 8 to 20 MB. Captures uploaded through the mobile app, web app and API were processed with the new pipeline, and the raw splat files could be exported for viewers such as SIBR and nerfstudio. Version 1.0.1 of the iOS app, released on 17 October 2023 and billed as "the big 1.0", brought Interactive Scenes and a "Luma v2.0" capture model to the phone. Later that month Luma released a Gaussian splatting version of its Unreal plugin, which Radiance Fields reported on 27 October 2023. It supported Unreal Engine 5.3, allowed NeRFs and splats in the same scene, and could crop, cull and merge splats; relighting worked only on parts directly lit by a light source. It was free, with commercial use allowed.

Genie and Series B (2023-2024)

In early November 2023 Luma opened Genie, a text-to-3D model, as a free research preview on Discord. A prompt produced four simple 3D models in about 20 seconds, downloadable as GLB files, and a refined version took about 20 minutes. Luma released Genie 1.0 with its Series B announcement in January 2024. It said Genie made a 3D object in under 10 seconds, as quad meshes with materials, on the web, in the iOS app and on Discord. Andreessen Horowitz led the round, and Luma named Amplify, Matrix, NVIDIA and South Park Commons as continuing backers. The company also hired Jiaming Song from NVIDIA's generative AI group as chief scientist and Matthew Tancik from Berkeley to lead applied research. Andreessen Horowitz credited Tancik with helping to create Neural Radiance Fields. TechCrunch put the round at 43 million US dollars and total funding at more than 70 million. A source told it the valuation was between 200 and 300 million US dollars. At that point Luma had 24 employees, its capture app had just over 2 million users and Genie had reached 100,000 users in four weeks. The capture app's App Store version history lists Genie's introduction on iOS in version 1.3.1, dated 17 January 2024.

Dream Machine and the move to video (2024-2025)

Luma released Dream Machine, a video generator that works from a text or image prompt, in June 2024. At its debut on 12 June 2024 it was free to try on Luma's website and produced five-second clips in about two minutes. The same month, a viral clip called "Monster Camp" made with Dream Machine drew attention for imagery resembling Pixar's Monsters, Inc. Luma said it had addressed the issue, which it believed came from someone uploading an image directly, not from its models scraping existing content from the internet.

Ray2 followed on 15 January 2025. Luma said it used ten times the compute of Ray1 and could make 5 to 10 second clips that could be extended to about 30 seconds. Luma made it available to paid Dream Machine subscribers. On 23 January 2025 Amazon Web Services added Ray2 to Amazon Bedrock and said it was the only cloud provider offering fully managed Luma models. Ray3, released on 18 September 2025, was described by Luma as a "reasoning" video model. It could output 10-, 12- and 16-bit HDR video in the ACES2065-1 EXR format and was also offered inside Adobe Firefly.

Series C and expansion (2025-2026)

On 19 November 2025 Luma announced a 900 million US dollar Series C at the U.S.-Saudi Investment Forum. HUMAIN, an AI company owned by Saudi Arabia's Public Investment Fund, led it. AMD's venture arm took part, along with existing investors Andreessen Horowitz, Amplify Partners and Matrix Partners. The two companies also said they would partner on Project Halo, a planned 2-gigawatt AI supercluster in Saudi Arabia, and on HUMAIN Create, an initiative for AI models trained on Arabic and regional data. Luma's own announcement said deployment of Halo would begin in the first quarter of 2026 and that the company had more than 130 staff.

Luma opened its first international office in London in December 2025, with Jason Day, a former Monks and WPP executive, as head of EMEA. It said it planned to create 200 roles there in 2026. In February 2026 it announced a Riyadh office and a partnership with Publicis Groupe Middle East & Turkey. Ray3.14 (January 2026) brought native 1080p output to the main Dream Machine workflows; Luma said it was four times faster and three times cheaper than Ray3 at 720p. Ray3.2 (June 2026) added frame-level control with up to 16 keyframes in a clip and clips of up to 20 seconds at 1080p. Luma said that, for the first time, the full control surface of the Ray model was also available as an API. In May 2026 CNBC ranked Luma 35th on its Disruptor 50 list, citing about 1 billion US dollars in funding and a valuation of about 4 billion US dollars (both PitchBook figures), with Amit Jain as CEO.

Products

Product	Introduced	Type	Notes
Luma 3D Capture (iOS app) and web app	2021 (app launch reported by TechCrunch), 1.0 in October 2023	3D capture	Turns iPhone video into NeRF captures, meshes and Interactive Scenes; exports to Blender, Unity and Unreal; version 1.3.14 released January 2026
Luma Unreal Engine plugin	April 2023 (alpha)	Game engine plugin	Free; renders Luma NeRFs in UE5; Gaussian splat support added in October 2023
Interactive Scenes	October 2023	Gaussian splat format and viewer	Real-time splat scenes on web and mobile, available through the API
Genie	November 2023 (preview), January 2024 (1.0)	Text-to-3D model	Sunset on 1 January 2026, according to the capture app's App Store listing; existing content can still be exported
Dream Machine	June 2024	Video and image generation platform	Described by Luma in 2026 as its flagship platform
Ray2, Ray3, Ray3.14, Ray3.2	2025-2026	Video generation models	Ray3 adds HDR EXR output; Ray3.2 adds multi-keyframe control and a full Ray API

XR work

Luma's capture tools make radiance field and Gaussian splat scenes from ordinary phone video. The App Store listing says no LiDAR or special capture equipment is required (capture needs an iPhone 11 or newer), and it describes exporting NeRFs and Gaussian splats to Unreal, Unity and other tools. The Unreal plugin's announcement listed games, virtual production, traditional graphics content and virtual reality experiences as intended uses. The app's version history also mentions an AR viewing mode for older scenes.

Luma also published luma-web, an npm library that rendered captures as Gaussian splats in WebGL and Three.js. Its documentation showed how to turn on Three.js XR rendering and add a VR button. It also warned that the result "will work best on PC VR, standalone VR tends to struggle with splats presently". The library's GitHub repository was archived in April 2025.

As of October 2026 the 3D capture app is still listed on the App Store, but Genie has been shut down and the company's recent announcements are about video and image generation and multimodal models.

See also

• Neural Radiance Fields
• Gaussian splatting
• Photogrammetry
• Polycam
• Volumetric video