AI & 3D generation · Advanced

NeRF (neural radiance field)

An AI model that learns a real scene from photos and shows it from new angles

Also called Neural radiance field

What is a NeRF (neural radiance field)?

A NeRF, short for neural radiance field, is a small AI model trained on photos of one scene that learns what color and density sit at every point in space. Once trained, it can render the scene from camera angles that were never photographed, with soft reflections and fine detail. It stores no mesh.

Think of it like this

It is like a person who has walked around a room so often they can picture it from any spot, without ever drawing a floor plan.

Where it fits in a 3D project

  1. Create
  2. Surface
  3. Rig
  4. Animate
  5. Render
  6. Export

Create: Building the shape: modeling, sculpting, scanning or generating it.

How a NeRF works

  1. 01You take photos or a video circling the scene.
  2. 02Software works out where the camera was for each photo.
  3. 03A small neural network learns the color and density of points in space until its renders match the photos.
  4. 04To view it, the network is asked what each camera ray passes through, pixel by pixel.

NeRF (neural radiance field) vs gaussian splatting

PropertyNeRF (neural radiance field)Gaussian splatting
Stores the scene asA neural networkMillions of soft colored blobs
Viewing speedSlow unless heavily optimizedReal time on many devices
Training timeLongerShorter
EditingHardPossible, but limited

Pick Gaussian splatting when you want to view a capture in real time; NeRFs are slower to view but remain common in research and some capture tools.

Where a NeRF is used

  • Realistic walk-throughs of real rooms, streets and buildings
  • Recording places as reference for film and games
  • Research into capturing scenes from photos

Beginner tip

Capture a NeRF as you would for photogrammetry: steady movement, lots of overlap, even light and no moving people. Blurry or moving frames turn into floating haze in the result.

Terms related to a NeRF

NeRF (neural radiance field) FAQ

What does NeRF stand for?
Neural radiance field. “Neural” because a neural network stores the scene, and “radiance field” because it records how much light, of what color, leaves every point in space in every direction.
NeRF vs Gaussian splatting: what is the difference?
Both learn a real scene from photos and show it from new angles. A NeRF stores the scene inside a neural network and is slow to view; Gaussian splatting stores it as millions of soft blobs and plays back in real time.
NeRF vs photogrammetry: which is better?
It depends on the job. Photogrammetry gives an editable textured mesh, good for games, film assets and printing. A NeRF looks more realistic, especially on shiny or thin details, but it is harder to edit or use outside a viewer.
Can you export a NeRF as a mesh?
Yes, many tools can pull a mesh out of a NeRF, usually with an algorithm called marching cubes. The mesh tends to be lumpy and loses the soft, view-dependent look, so it often needs cleanup or retopology.

Updated October 29, 2026 · Browse all 3D terms