Relighting News
The latest Relighting coverage curated for game developers.
arXiv cs.GR details GS-PI, a method for turning Gaussian-splat assets into fully relightable PBR content without per-scene joint optimization. The approach separates appearance decomposition from geometry, then uses diffusion and …
LightBridge aims to make relighting 3D Gaussian Splatting assets a one-pass process instead of a per-scene optimization job. The system uses a feed-forward generative pipeline plus a new relighting dataset …
ExMesh++ turns multi-view photos into editable UV-PBR mesh assets, aiming to skip the usual mesh extraction and texture-baking detour. The staged pipeline adapts topology with vertex split/merge operations, then locks …
Luce is a new single-image-to-3D pipeline built around relightable Gaussian assets, aiming to preserve geometry, materials, and fine surface detail in one representation. It targets PBR-ready output with albedo, metallic-roughness, …
A new intrinsic decomposition method tackles a long-standing misattribution problem in point-based 3D scene reconstruction. Intrinsic PAPR replaces translucent volume aggregation with proximity-aware point rendering so each surface point can …
IRGS++ pushes Gaussian splatting beyond pretty rasterized views by adding transport-aware inverse rendering for visibility and indirect light. The method targets inter-reflections, glossy and metallic materials, and novel relighting, aiming …
OmniX turns panoramic generation into a pipeline for graphics-ready 3D scenes, aiming beyond pretty views toward geometry, lighting, and PBR materials. The system repurposes pretrained 2D flow-matching priors with cross-modal …
A new inverse-rendering method uses rigid object motion to separate material from lighting more reliably. By tracking and reconstructing moving objects, the pipeline gets multiple light-surface interactions from a single …
LightCrafter tackles video relighting by feeding a diffusion model a PBR-rendered proxy instead of the raw footage. That shift aims to keep long-form shots temporally stable while giving artists more …
A new neural texturing method targets a long-standing pain point in photogrammetry-style pipelines: baked lighting in captured textures. DANTE-W reconstructs diffuse albedo from unstructured image sets, aiming to preserve material …
A paper from George Drettakis and collaborators extends intrinsic decomposition to 3D Gaussian splats, letting teams separate albedo from shading and edit texture without baking in lighting. The practical win …
AEGIR pushes relightable Gaussian Splatting closer to usable indoor lighting by modeling real area emitters instead of faking them as points or environment maps. That matters for anyone trying to …
For graphics teams working with 2D Gaussian splatting, the big change is that DR-GS stops baking lighting into the reconstruction. It separates geometry, illumination, and material so deformed objects can …
A new arXiv paper tackles single-image human material estimation, a hard problem for relighting and digital human workflows because lighting, shape, and reflectance are tangled together. HAFMat uses hybrid priors …
A new diffusion-editing method adds a residual image embedding to improve both identity preservation and edit precision. For teams experimenting with generative tools, the interesting part is that it reduces …
A new rendering pipeline claims to make captured 3D scenes both editable and photorealistic by combining 3D Gaussian ray tracing with a lightweight neural renderer. The practical angle for graphics …
For graphics programmers, the interesting bit is that this paper moves Gaussian splatting beyond per-primitive BRDF fitting. MaterialClusterGS uses a shared palette of material prototypes plus a continuous material field, …
A new inverse-rendering approach for 3D Gaussian fields tries to close a long-standing gap between training and final rendering. Instead of optimizing materials in screen space, it keeps forward rendering …
For teams working on reflective assets, this paper tackles a real pain point: 3D Gaussian Splatting is fast, but it struggles to produce reliable geometry and relightable surfaces on shiny …
Disney Research’s new head-relighting model removes a big capture bottleneck: you no longer need OLAT footage for every subject. It can learn from ordinary single- or multi-view images, then generalize …