Skip to content
Atlas World Model
Menu

Video to 3D world generator: reconstruct scenes from ordinary footage

Walk through a room with your phone and get an explorable 3D world back. Video input is an Atlas capability; this page will unlock when World Labs opens the Atlas API. Until then, extract a few frames and use the image-to-3D tool.

Estimated cost

Coming when the API opens. Generation is not live on this site yet; estimates use the current pricing model.Join the waitlist

How it works

  1. Record 10–60 seconds of steady footage moving slowly through the space.
  2. Upload the clip (MP4/MOV, up to 500 MB); frames are sampled automatically.
  3. The world model reconstructs the scene as a Gaussian splat and estimates camera paths.
  4. Export the world, or render a new camera move — including bullet-time freezes — through it.

Input requirements

  • Slow, smooth camera motion with plenty of overlap between frames.
  • Avoid moving people or vehicles; static scenes reconstruct best.
  • Landscape orientation and 1080p or higher recommended.

Output formats

  • Gaussian splat (.spz, .ply)
  • Textured mesh (.glb) — Pro plan and above
  • Re-rendered camera-move video (.mp4)

Preview every output in the free Gaussian splat viewer before downloading.

Pricing

Generation is credit-based. World Labs sells credits at $1 for 1,250 credits, and a Marble 1.1 world costs 1,500 credits, about $1.20 (as of 2026-09-03); PLY splat export is free. This site's plans start at $19/month. Full breakdown on the pricing page.

Example worlds

Sample worlds generated through the Marble World API will be added here as the gallery grows. In the meantime, browse World Labs' public Marble gallery for reference and open any exported splat in the viewer above.

Last updated 2026-09-03

Frequently asked questions

Why is video input not available yet?

Video-to-world is an Atlas feature; the public Marble API accepts images, panoramas and text. This tool activates when Atlas API access opens.

Can I make a bullet-time effect from a phone video?

That is the goal: reconstruct the scene, then render a new frozen-moment camera orbit through it. Expect this after Atlas access opens.

What about robotics real-to-sim?

Video capture of a real workspace is a natural way to build simulation twins; see the robotics use-case guide for the workflow.

What can I do today?

Extract 3–4 well-spaced frames from your video and use the image-to-3D tool, or view existing splats in the free viewer.

Get notified when video to 3d world generator goes live

Be first to know when Atlas opens beyond partners, and when image-to-3D and text-to-3D world generation goes live on this site.

No spam. One email when Atlas access or the API opens on this site, plus occasional major updates.