Improving energy conservation and multi-scattering for rough materials in UE5

This project integrates a Kulla–Conty BRDF into Unreal Engine 5 to improve specular energy preservation and multi-scattering approximation for rough surfaces in real-time rendering.

It demonstrates my experience in physically based rendering, Monte-Carlo integration, BRDF modeling, LUT precomputation, and Unreal Engine shader system customization.

Unreal Engine 5’s default GGX-based BRDF models primarily single-scattering and applies limited energy compensation.
At medium-to-high roughness, noticeable specular energy loss still occurs, resulting in dim highlights and less physically plausible materials.

This project implements a Kulla–Conty BRDF model to approximate:

• Multi-bounce microfacet scattering
• Improved energy conservation
• More physically realistic rough surface reflections

The model is evaluated in real time using precomputed lookup tables (LUTs) and a custom HLSL material node inside UE5.

• Monte-Carlo integration of Kulla–Conty BRDF
• Split-sum approximation for real-time evaluation
• Multi-scattering energy compensation
• Offline precomputed LUT pipeline
• Custom HLSL integration into UE5 material system
• Visual comparison under controlled lighting

Standard GGX BRDF only models single scattering, which leads to:

• Specular energy loss
• Darkened highlights on rough materials
• Non-physical appearance under strong lighting

The Kulla–Conty BRDF approximates multiple microfacet bounces, redistributing lost energy back into the specular response.

This project explores:

How much visual improvement can physically-based multi-scattering bring to real-time UE5 materials?

We follow the Kulla–Conty microfacet framework with:

The specular BRDF integral is decomposed into:

• A-term: non-Fresnel component
• B-term: Fresnel-weighted component

Stored in a 2D LUT:
E_mu(N·V, roughness)

We additionally compute:

• E_avg: average energy loss of single scattering

This allows reconstructing the multi-bounce contribution at runtime.

Final evaluation requires only:

• 2 LUT samples
• Fresnel evaluation
• Simple normalization math

No runtime Monte-Carlo sampling is required.

Two 256×256 textures are precomputed on CPU:

• Hammersley low-discrepancy sampling
• GGX importance sampling
• 1024 samples per texel
• Smith masking-shadowing
• Schlick Fresnel approximation

• X: N·V
• Y: Roughness

The generated LUTs are stored as linear PNG textures and imported into Unreal Engine.

• sRGB: OFF
• Compression: VectorDisplacement
• Filter: Bilinear / Trilinear

1. Compute N·V using CameraVectorWS · PixelNormalWS
2. Use (N·V, Roughness) as LUT UVs
3. Sample:
   • E_mu → A & B terms
   • E_avg → average energy
4. Evaluate final BRDF using a Custom HLSL Node

The output replaces UE5’s default specular BRDF response.

• Metallic spheres from roughness 0 → 1
• Directional light
• Top: UE5 default BRDF
• Bottom: Kulla–Conty BRDF

Results:

• Low roughness: nearly identical
• Medium roughness: stronger, more stable highlights
• High roughness: better energy preservation

• Controlled lighting environment
• Identical materials
• Side-by-side comparison

Results:

• Improved brightness consistency
• Better grazing-angle response
• Clearer energy retention at mid roughness

• UE5’s default BRDF already applies partial compensation
• Full Kulla–Conty BRDF provides more physically accurate energy behavior
• Improvements are most noticeable at medium roughness
• Suitable for:
  • Rendering research
  • Shader and material system development
  • Physically-based lighting studies

• Evaluated only on simple test scenes
• Performance overhead not deeply profiled
• Visual differences are subtle in low-complexity materials
• More complex layered materials may benefit more

• Designed LUT precomputation pipeline
• Implemented Monte-Carlo GGX integrator
• Generated E_mu and E_avg textures
• Integrated custom BRDF into UE5 material system
• Built evaluation scenes and comparison tests

• Kulla, C. & Conty, A.
  Revisiting Physically Based Shading at Imageworks
• Epic Games — Unreal Engine 5 Rendering Documentation
• UE5 Fab Parametric Cornell Box

Zihan Wang (王滋涵)
Computer Graphics / Rendering / XR

GitHub: https://github.com/ZihanWG Portfolio: https://zihanwg.github.io/portfolio.github.io / https://zihanwg.github.io/ZihanW.github.io/

Technical write-up:
https://zihanwportfolio.wordpress.com/2025/05/06/integrating-kulla-conty-brdf-for-real-time-pbr-in-ue5/