Creating worlds from video
Upload short videos to create immersive 3D worlds with rich spatial information. Videos provide multiple perspectives of your space, allowing the AI to understand depth and spatial relationships better than a single images alone.Getting started with video upload
Drag a short video of a static space into the omnibox; it will expand with 2D input mode, which allows you to start a generation by clicking on Create. For video file specifications, see Prompt guidelines →Best practices for video capture
Most artifacts in the generated world trace back to the source video, so the biggest improvements come from how you record.Camera movement guidelines
- Rotation focus: Rotate the camera to cover a large viewing angle
- Avoid motion blur: Use steady, controlled camera movements to avoid excessive motion blur
- Wide coverage: Aim to capture 180° to 360° of the space
- Continuous shot: Record one uninterrupted take of the space
Camera settings
- Fixed focal length: Avoid zooming in or out during recording
- Fixed exposure: Avoid changing exposure during recording
Troubleshooting common issues
Artifacts or blur in the generated world
- Re-record with steady motion: Motion blur and shaky footage carry into the world as smeared or noisy geometry.
- Keep the space static: Avoid moving people or objects; the model reconstructs a static scene best.
- Lock focal length and exposure: Zooms and auto-exposure shifts during the take confuse depth estimation.
- Export at full resolution: Choose the 2M-splat SPZ or PLY rather than the low-res (500k) option for finer detail. See Export file specs.
- Clean up specific spots: Use Pano edit for targeted fixes, or generate a Variation to get a cleaner result.