Google Patents a Way to Fix Lighting Across Every Angle of a 3D Scene
When you stitch photos together into a 3D scene, some angles look washed out and others look too dark. Google's new patent attacks that problem at the source by baking correct exposure into the 3D model itself, not patching it up image by image afterward.
How Google's exposure fix works in 3D reconstructed scenes
A photographer shoots a room from a dozen angles. Some shots are a little overexposed near the window; others look dim in the corner. When software tries to knit those photos into a navigable 3D space, those mismatched exposures show up as jarring pops of brightness as you move through the scene. That's a well-known headache in 3D reconstruction.
Google's patent describes a system that solves this by teaching the 3D model itself what the correct brightness should be at every point in space, not just in each flat photograph. Instead of adjusting brightness photo by photo after the fact, the model learns exposure as part of building the scene, so every angle you look from stays consistent.
For you, the practical result is 3D scenes and AI-generated views that don't flicker or shift in tone as the virtual camera moves. Think of it as automatic color correction that lives inside the model rather than on top of it.
receiving, using a first model, a brightness at a three-dimensional location of a scene; receiving a first perspective of the scene; and generating, using a second model, a second perspective of the scene based on the first perspective of the scene and the brightness at the three-dimensional location of the scene.
Translation: AI models take lighting data from one angle to render a new view.
How the neural field assigns brightness per 3D point
The patent describes what Google calls a neural exposure field, an extension of a technique called a neural radiance field (a model that reconstructs a 3D scene from 2D photographs and can then render that scene from any new camera angle).
The key addition is a second model that learns a single optimal exposure value for each point in 3D space. Exposure here means roughly how bright or dark that point should appear when lit and photographed under normal conditions. Rather than correcting brightness in 2D (per image) after the scene is built, the exposure correction is baked into the 3D representation from the start.
The patent also describes how this exposure model is trained together with a grid-based radiance field (a faster, memory-efficient way to store scene data using a 3D grid rather than a pure neural network). A mechanism called latent exposure conditioning lets the model adjust for the unique camera settings of each input photo during training, then strip those camera-specific quirks out of the final scene.
- Input: photos from multiple angles, each with its own camera exposure setting
- Process: a neural network learns both scene geometry and per-point brightness simultaneously
- Output: a 3D scene where any new rendered view has consistent, well-exposed color throughout
Techniques are directed to neural exposure fields, an improved technique for reconstructing scenes with 3D consistent and high-quality appearance that shows optimally-exposed colors while being consistent in 3D.
Translation: The system uses neural exposure fields to keep colors balanced in 3D.
What this means for 3D photography and AI-generated views
The visible payoff is simple: 3D scenes that don't shift in brightness as the virtual camera moves. Today, many 3D photo tools and AI video systems handle exposure photo by photo, which means moving through a reconstructed space can feel like someone is toggling the lights. This approach removes that artifact.
For people using AR tools, 3D scanning apps, or AI image generators that rely on scene reconstruction, Google's steady investment in neural rendering could translate into output that looks more like a single, coherent photograph and less like a patchwork of mismatched snapshots. The impact would be most obvious in challenging real-world conditions: scenes with bright windows, outdoor sun, or mixed indoor and outdoor lighting.
This is the 47th Google filing we've tracked in AI photo editing since May, following one on learning camera angles and another on gesture-based image edits.
Anyone who has explored a 3D map or walked through a photo-realistic virtual space has noticed that unsettling flicker when the image seems to brighten or darken as the camera moves, even though nothing in the scene actually changed. That visual glitch erodes trust in what you're seeing, and this patent is specifically designed to prevent it.
Google's approach bakes exposure correction into the three-dimensional model of a scene rather than patching each individual photo after the fact. The result is that brightness stays stable and consistent no matter which direction you look, the way lighting in the real world actually behaves.
Most users will never know this fix exists, which is precisely the point. You notice it only when it's absent: a virtual tour that feels slightly off, a street-view transition that looks wrong. Removing that distraction is a quiet but concrete improvement to how believable these experiences feel.
There are more where this came from
We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.
The drawings
5 drawing sheets from US 2026/0268585 A1 · click any drawing to enlarge
Want this weekly breakdown for a company we don't cover? Patentlyze Pro →