Skip to content

A Diffusion Approach to Radiance Field Relighting using Multi-Illumination Synthesis

Relight a radiance field from multi-illumination synthesis

Visit official site

What Radiance Field Relighting is

The method is described as a way to create a relightable radiance field using priors extracted from 2D image diffusion models. It converts multi-view data captured under one lighting condition into a multi-illumination data set.

What you can do with it

  • Augment single-illumination data with a 2D diffusion model
  • Control light direction through a 2D relighting network
  • Tune auxiliary feature vectors for multi-view consistency

Application scenarios

  • Create lighting effects for 3D scenes in film production
  • Enhance virtual settings in virtual reality and game development
  • Simulate building exteriors under different lighting in architectural visualisation

Pros & cons

✓ What we like

  • The approach does not rely on exact geometry or surface normals
  • Complex geometry and reflective BRDF scenes are named as suitable

! What to watch out for

  • The workflow starts from single-illumination multi-view data
  • Synthetic relighting inputs are treated as potentially inaccurate

FAQ

What does the method start from?

The steps begin with multi-view data captured under a single lighting condition.

How is the field represented?

The material describes a relightable radiance field represented with 3D Gaussian splats.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review