Real Time Rendering News
The latest Real Time Rendering coverage curated for game developers.
A new global illumination method packs light transport into a 13D Gaussian mixture over position, direction, normals, and material. The result is millisecond-scale, view-independent rendering with far less memory than …
AMD GPUOpen details a one-step latent diffusion approach for real-time global illumination that aims to keep lighting stable without iterative denoising passes. For graphics programmers, the pitch is better GI …
A new rigging approach for Gaussian splats uses differentiable skinning weights to avoid the holes and stretching artifacts that show up when points are only moved rigidly. The method stays …
Neural texture compression can beat BCn on size, but shader-side divergence has been the tax. A new decoder design cuts that divergence by 25% to 52% and reaches up to …
A new water-simulation coupling scheme tackles one of graphics’ long-standing seam problems: mixing cheap 2D wave models with localized 3D fluid solvers. By using a nonlinear, dispersive Hamiltonian surface model, …
A new film-emulation pipeline predicts a 3D LUT from a single reference frame, aiming to match unseen analog looks without paired training data. It also sidesteps the usual LUT-bank collapse …
A new texture-compression approach targets mipmapped SVBRDFs with shared 2D Gaussians, aiming to cut storage without giving up random access. The method, called GTC, separates spatial footprint from per-map values …
A new radiance-field rendering method replaces convex splats with deformable triangles that can bend inward or outward. That lets curved and concave surfaces be represented with fewer primitives while keeping …
A new real-time GI technique, Split Radiance Cascades, targets full 3D diffuse lighting with sparse world-space probes instead of screenspace tricks. It combines a hashmap-backed probe layout with ray splitting …
A new splatting pipeline promises up to 5x faster rendering than 3D Gaussian Splatting while keeping state-of-the-art visual quality. It bakes high-frequency detail into a compact texture atlas, then uses …
Gaussian splatting is moving beyond static capture and into virtual production workflows. DOSENMUTTER used Unreal Engine to turn a childhood bedroom into a giant-scale film set, including a human-sized can, …
Real-time strand hair just got a practical boost: a new deferred software rasterization pipeline renders hair meshes with minimal hardware support. The approach targets far-field hair at even one sample …
A new ASR approach, QuADA-GS, tries to make arbitrary-scale upscaling practical by allocating compute only where image detail actually needs it. Instead of treating every pixel uniformly, it routes a …
FiCA turns a single portrait into a drivable, photoreal 3D Gaussian avatar, cutting out per-person test-time optimization. The pipeline blends human-centric vision foundation models with diffusion and a mesh-refinement stage, …
A new rendering pipeline can push radiance fields onto light field displays in real time, hitting 200+ FPS at 512p across 45 views on a Looking Glass panel. For graphics …
Dynamic 3D Gaussian splatting is getting a practical speedup: this paper claims 10-minute training and 700+ FPS on a single RTX 5090 for Neural 3D scenes. The key trick is …
Disney Research’s RenderFlow replaces iterative diffusion with a deterministic single-step neural renderer, aiming for near real-time output without the usual stochastic flicker. For graphics teams, the interesting part is the …
A new real-time hair reconstruction method targets one of rendering’s hardest edge cases: strand aliasing. It rebuilds hair coverage, tangent, and position from heavily undersampled raster inputs, then feeds that …
A real-time lighting system for procedurally generated isometric terrain tries to fake 3D light interaction from 2D map data, which is the interesting part for graphics and tools folks. The …
HumanGS reconstructs and animates people in real time from sparse multi-view RGB and SMPL-X poses using feed-forward Gaussian splatting. It drops depth supervision, UV maps, and per-pose re-inference, while still …