✨ 3D Gaussian Splatting

The revolution in real-time 3D — better than NeRF, without neural networks

What is Gaussian Splatting?

3D Gaussian Splatting (3DGS) is a new method for 3D reconstruction from photos/videos. Instead of meshes or neural networks it uses millions of tiny ellipsoids ("Gaussians") that together represent a scene.

✅ Advantages

  • Real-time rendering (60+ FPS)
  • No neural network required
  • Editable in Blender
  • Realistic view synthesis

⚠️ Drawbacks

  • High VRAM demand (8 GB+)
  • Large files
  • No real mesh
  • Still young (2023)
📖 Get started right away: The Getting Started & Software Guide 2026 shows the fastest route to your first splat (phone, free) and compares all current tools. The Workflow Guide explains the full process for both techniques.

NeRF vs. Gaussian Splatting

NeRF (Neural Radiance Fields) was the predecessor — impressive, but slow to render because every pixel has to pass through a neural network.

Gaussian Splatting renders 100–1000× faster because it uses classic rasterization instead of ray marching through a network.

Capture for Gaussian Splatting

Video vs. Photos

Recommendation: video in 4K, 30 fps with slow, steady movement.

Important during capture

⚠️ Hardware requirement: Training needs an NVIDIA GPU with at least 8 GB of VRAM. 12 GB+ recommended for larger scenes.

Training Workflow

  1. COLMAP: compute the camera positions from the images
  2. Training: optimize the Gaussians (15–30 minutes)
  3. Export: .ply or .splat file
  4. Viewer: SuperSplat, Unity, Unreal or web
💡 Easy start: Luma AI does all of this in the cloud — upload a video, done. Free to try.

Software

More details in the Software Comparison →