Skip to main content

​
Welcome to Splat3D

Splat3D turns your photos and videos into interactive, shareable 3D scenes. Upload a video walkthrough, and get back a fully reconstructed 3D Gaussian Splat you can embed, explore, and share with a link.

Quickstart

Create your first 3D scene in 5 minutes.

API Reference

Full endpoint documentation with interactive examples.

Quality Presets

Choose the right speed/quality tradeoff for your use case.

Training Parameters

Fine-tune every aspect of the reconstruction pipeline.

Pricing & Credits

Pay-as-you-go credits, the price of a scene at each quality tier, and credit packs.

​
What Splat3D does

  1. Upload a video or a zip of photos, optionally with ARKit camera poses from your phone
  2. Reconstruct camera poses using GPU-accelerated structure-from-motion (GLOMAP)
  3. Train a 3D Gaussian Splat model using gsplat on L4, A100 or H100 GPUs, by preset
  4. Compress the result into a web-optimized SOG format for instant loading
  5. Serve the 3D scene through a shareable link or embed it in your app
The entire pipeline runs on serverless GPUs — you don’t need to manage any infrastructure.

​
Key features

  • Configurable quality presets — from fast previews (3K iterations, ~8 min) to ultra quality (30K iterations, ~45 min)
  • Full parameter control — override training iterations, resolution, SfM matching, and more
  • Multiple input types — video files, photo archives, and ARKit captures from iOS
  • Web-optimized output — SOG (compressed Gaussian Splat) format loads instantly in any browser
  • Pre-supplied poses — skip structure-from-motion entirely if you already have a COLMAP model
  • Quality metrics — get the gaussian count back with every completed scene, and holdout PSNR and SSIM by default on every preset but fast

​
Who it’s for

  • Developers building 3D experiences into their apps
  • Real estate platforms creating virtual walkthroughs
  • E-commerce teams building product viewers
  • AR/VR creators generating 3D content from real-world captures
  • Researchers working with Gaussian Splatting at scale

​
Base URL

All API requests go to:
The API is currently in beta. We’ll do our best not to make breaking changes, but the advanced training parameters may evolve between versions.