VFX Atlas
H 320 / 3212020s

Neural Relighting

2020s; Google’s Portrait Light introduced in 2020

Typeface · Dhurjati

AI illustration: Neural portrait relighting with warm directional source.
AI illustration: Neural portrait relighting with warm directional source.Illustration generated for the Atlas (AI)View source ↗

Prompt

Apply Neural Relighting with a warm directional key light from the upper left and cool, soft fill from the right. Redirect cast shadows consistently, shift specular highlights to match surface geometry, and preserve natural material textures without changing viewpoint. Keep the original subject and composition.

Paste it into LetzAI with your own image to apply the effect ↗

Identify

A photographed face or scene gains redirected shadows, shifted highlights, or a new lighting color. The viewpoint stays fixed while the illumination appears to change.

Principle

A neural network1 predicts how image pixels should change under different lighting, using learned patterns of shape, materials, and illumination. Some methods estimate surface geometry2 explicitly; others generate the altered appearance directly.

Example

Google: Portrait Light, 2020.

Try it out

Relight one portrait from opposite sides, then compare the nose shadow, cheek brightness, and reflections in the eyes.

Background

Neural relighting became more accessible during the 2020s, building on earlier computational photography and computer graphics research. Google’s Portrait Light offered a prominent example in 2020, adding adjustable illumination to portraits after capture. Training can use photographs under controlled lights, rendered scenes, or other image collections. The central difficulty is separating surface reflectance3 from illumination: a dark patch might be paint, shadow, or both. A single photograph rarely provides enough evidence for a unique solution. The network therefore makes learned guesses, which can produce convincing but physically inconsistent results.

Distinction

Unlike a global exposure adjustment, neural relighting can change local shading according to inferred shape. It changes the image after capture, not the original exposure.

Aftermath

Neural relighting shows that convincing illumination can be synthesized without proving how the original scene was lit.

Further examples · 7 images

AI illustration: Multi-directional light source prediction on human face
02 AI illustration: Multi-directional light source prediction on human faceThree panels show an elderly man under varying light directions that dramatically alter the appearance of his facial features. The consistent geometry across frames indicates a neural model predicting light transport over surface textures.Illustration generated for the Atlas (AI)
AI illustration: Synthetic cinematic color relighting of a portrait
03 AI illustration: Synthetic cinematic color relighting of a portraitA split comparison shows a man in neutral lighting transformed with a dual-tone blue and orange cinematic light scheme. The neural network has accurately simulated how these colored light sources wrap around his features and clothing.Illustration generated for the Atlas (AI)
AI illustration: Shadow redirection on a classical plaster sculpture
04 AI illustration: Shadow redirection on a classical plaster sculptureA plaster bust is depicted with shifted shadows that realistically follow the complex curves of the sculpture. This provides a clear look at how neural relighting handles non-organic materials and cast shadows.Illustration generated for the Atlas (AI)
AI illustration: Still life relighting with metallic surface highlights
05 AI illustration: Still life relighting with metallic surface highlightsA still life arrangement featuring a teapot and fruit shows a transition from flat lighting to a warm, low-angled light. The reflections on the metallic surface change appropriately, demonstrating material-aware light estimation.Illustration generated for the Atlas (AI)
AI illustration: Environmental scene relighting with complex shadow projection
06 AI illustration: Environmental scene relighting with complex shadow projectionAn interior scene with a chair and plant gains long, golden-hour shadows cast across the wall and rug. The complex silhouettes of the leaves show the high-fidelity shadow synthesis typical of neural methods.Illustration generated for the Atlas (AI)
AI illustration: Neural light manipulation on complex fur textures
07 AI illustration: Neural light manipulation on complex fur texturesA domestic cat is shown with a new light source added to the left, creating a rim light effect on its fur. The technique successfully navigates the difficult task of relighting fine, fibrous textures like animal hair.Illustration generated for the Atlas (AI)

Question

Why might a relit face look plausible while its glasses look wrong?

Show possible answer +

Skin shading and glass reflections respond differently to light. The network may predict facial shading well but misjudge reflective or transparent surfaces.

Sources