Jessie A Ellis
Feb 05, 2026 14:34
Small dev team uses PBRFusion AI to convert thousands of legacy textures into modern PBR materials, cutting years off Painkiller RTX remaster timeline.
A small modding team has demonstrated how generative AI can compress years of texture work into weeks, rebuilding the 2004 shooter Painkiller with path-traced visuals across 35 levels using NVIDIA’s RTX Remix toolkit and custom AI models.
The project showcases a production pipeline that processed thousands of legacy textures into modern Physically Based Rendering materials—work that would have been impossible for a small team using traditional methods.
AI Handles 80% of the Grind
McGillacutty, the project’s environment and material lead, puts it bluntly: manually rebuilding thousands of materials across 35 levels with minimal texture reuse wasn’t happening. Enter PBRFusion, an open-source model created by team member NightRaven specifically for RTX Remix workflows.
The model batch-generates base color, normal, roughness, and height maps from legacy textures. Quinn Baddams, team lead and founder of Merry Pencil Studios, estimates AI automation eliminated roughly 80% of repetitive work, freeing the team to focus on the 20% requiring artistic judgment.
“PBRFusion was always intended to be a tool, not a drop-in replacement,” NightRaven explains. The model went through three major iterations and over 1,000 hours of development before reaching the version used in production.
Where AI Falls Short
The team found clear limits to automation. Metallic materials were largely hand-crafted. Glass, transparent surfaces, and skin required custom values and maps—particularly for subsurface scattering effects. Texture atlases, where single images contain multiple unrelated surfaces, confused the AI models entirely.
Roughness maps proved especially tricky. “AI-generated roughness often requires adjustment to achieve physically accurate results,” Baddams notes. “Correct values can be very specific.” The team cross-referenced real-world PBR materials to validate their outputs.
Hero materials received additional treatment using CC0 PBR blending, procedural workflows in InstaMAT Studio, and manual painting. The hybrid approach—AI baseline plus human refinement—became their standard operating procedure.
Path Tracing Exposes Everything
Full-scene path tracing fundamentally changed the material workflow. Properties like roughness, reflectivity, and wetness became far more visible than traditional rendering ever revealed. The original game’s baked shadows, which added contrast in 2004, now created physically impossible lighting responses.
The solution: strip baked lighting from source textures, then reintroduce contrast through roughness variation, stronger normal maps, and controlled self-shadowing. RTX Skin technology enabled subsurface scattering on characters—light genuinely scattering through surfaces rather than simple highlight tricks.
“To my knowledge, this level of ray-traced subsurface scattering hasn’t been available to game developers in a practical, real-time way,” Baddams says. “It was previously limited to offline rendering.”
Implications for the Industry
The project arrives as the broader Painkiller franchise sees renewed activity. 3D Realms announced a modern reimagining in March 2025, targeting an October 2025 release for PlayStation 5, Xbox Series, and PC. The timing creates an interesting parallel: official studio resources versus small-team AI-augmented workflows achieving comparable visual ambitions.
For developers considering similar approaches, the team recommends starting small—capture a single scene, apply basic PBR materials, iterate with path tracing to understand material-light relationships before scaling up.
NightRaven is already finishing the next PBRFusion version. NVIDIA will showcase related RTX neural rendering advances at GDC, where VP John Spitzer will present path tracing and generative AI workflow innovations. The tooling gap between AAA studios and indie teams continues narrowing.
Image source: Shutterstock
