What Luma AI is
Luma AI started with a deceptively simple idea: point your phone at something, walk around it, and get back a 3D version of it that actually looks real. No laser scanner, no photogrammetry rig, no manual cleanup. Built by Luma Labs in Palo Alto, the platform uses neural radiance fields (NeRF) and Gaussian splatting to reconstruct full scenes from ordinary photos and video, then lets you view, share, or export them. Over time the company has widened its scope into generative video, but the 3D capture heritage is what put it on the map and remains a reference point in the radiance-fields community.
It sits in the AI 3D model generator category, though it is worth being precise about what kind of 3D it produces. Luma is strongest at capturing real places and objects as photoreal scenes, rather than spitting out a clean, game-ready hard-surface mesh from a text prompt.
How the capture pipeline works
The workflow is genuinely approachable. You record a short video or shoot a set of photos of an object or environment with any smartphone, upload it to Luma's cloud, and the service handles reconstruction on its own servers. Within minutes to a few hours, depending on scene complexity and queue, you get back a processed Gaussian splat scene. Splats render the world as millions of tiny coloured points rather than polygons, which is why reflections, soft edges, and translucency look so convincing compared with traditional mesh scanning.
Finished captures live in Luma's web gallery where they can be embedded, shared on social, or exported. Point-cloud data can come out as PLY files for use in external 3D tools, and there are pathways to bring captures into Unreal Engine and Unity through dedicated plugins, which is where game and virtual-production teams pick it up.
Standout features
Gaussian splatting and NeRF capture
This is the core. Smartphone footage in, photoreal interactive scene out. The splat representation preserves lighting and reflections that polygon-based scans usually flatten or lose entirely.
Flythroughs and stabilised video
Luma can generate smooth camera flythroughs of a captured scene and produce highly stabilised video, which is useful for real-estate walkthroughs, product showcases, and establishing shots.
Genie text-to-3D (legacy)
Genie was Luma's text-to-3D model generator that produced quad-mesh assets from prompts in seconds. Be aware that Genie was sunset on January 1, 2026; existing users could still export their content, but it is no longer a path for new mesh generation. If clean prompt-to-mesh output is your priority, a dedicated mesh generator is a better fit today.
Generative video and agents
Luma's current public direction leans heavily into cinematic AI video, including its Ray models and Dream Machine, plus creative agents aimed at agencies and production teams. This is a different product surface from 3D capture, but it shares the same account and underlying research.
Who it is for
Luma fits creators and teams who need to bring the real world into 3D fast. That includes visual-effects and virtual-production crews capturing locations and props, real-estate and architecture professionals making interactive walkthroughs, product marketers building 360-degree showcases, and hobbyists who just want to scan something interesting. Game studios use the Unreal and Unity export paths for environment reference and background set dressing.
It is less suited to anyone who needs tidy, low-poly, riggable character meshes straight from a text prompt. Splats are beautiful for viewing but are not a drop-in replacement for hand-built game assets, so factor that into your pipeline.
Pricing and access
Luma has offered a free tier that covers basic capture and processing with sharing through the web gallery, which is part of why it spread so widely. Paid and enterprise options add things like API access and priority processing for commercial use, and capture-via-API has been offered on a per-capture basis. Pricing shifts as the product evolves toward video, so check the current plans on Luma's site before committing for a project.
Verdict
For turning real spaces and objects into photoreal 3D with nothing but a phone, Luma AI is one of the most accessible options out there, and the splat quality holds up. Just go in clear-eyed about what it does best: capture and scene reconstruction, not prompt-to-clean-mesh asset generation, especially now that Genie has been retired. Used for the right job, it removes an enormous amount of friction from getting reality into a 3D workflow.
