In 2025, PBR is still the production material model. What changed is how maps are made, streamed, and lit. If you chase every product announcement, you will rebuild a library every quarter. If you watch the actual pipeline shifts, you can keep a CC0 metalness/roughness shelf and adopt only the pieces that save time.
The map set did not get replaced
Albedo, roughness, normal, and metalness remain the default in Blender, Unreal, Unity HDRP, and most DCC exporters. Specular/gloss still exists in some film VFX graphs, but new game and archviz work is metalness. Learn the slots from the PBR textures complete guide and ignore “new PBR” branding that is just a UI skin.
What did grow: optional maps on the same shader. Add them when the surface needs them, not because a pack includes empty 4K files:
- Displacement or height for close-up stone and wood.
- Coat and sheen for car paint, varnish, and fabric.
- Transmission only for glass and thin plastics.
Authoring: scans, AI, and hybrid libraries
Photogrammetry and material scanners still produce the most believable tiling surfaces. AI tools now fill gaps: upscaling, seam hiding, generating a roughness from a photo, or proposing a variation. The useful 2025 habit is hybrid: start from a real scan, then use AI for variants, not the other way around. Generated-only materials still fail energy ranges and tile badly under GI.
Free CC0 scans remain the cheapest honest base. Locker3d textures are free CC0; you can run them through an upscaler or a variation tool and ship the result. Confirm that in CC0 what you can and cannot do. Do not run a paid, non-transferable pack through a public generator if the license forbids derivative cloud processing.
Streaming and resolution: 4K is not the default
Virtual textures, Nanite-style mega geometry, and streaming samplers make 4K and 8K more practical on hero worlds. They do not make 8K mandatory. Texel density on screen still wins. Follow the resolution guide: 2K for most real-time, 4K for close-ups, 1K for clutter. A streamed 8K wood that tiles every 80 cm still looks like a stamp.
Build the library at 2K, keep 4K next to hero wood and metal only. Browse the Locker3d catalog at the size you will actually bind.
Lighting: IBL plus RT/GI as the new lookdev floor
In 2025, approving a material under a single gray studio HDRI is not enough if the shot uses Lumen, path tracing, or hardware RT. Lookdev under IBL and at least one GI or RT bounce. Practical IBL steps: HDRI and PBR lighting. Trends in lighting do not change how you author albedo; they only punish baked lighting faster.
What to do this year, in order
- Standardize on metalness/roughness, OpenGL or DirectX normals documented per project, sRGB only on albedo.
- Keep a CC0 core library you can legally modify. Add paid or custom scans for unique heroes.
- Adopt AI as a retouch and variation layer, not as the only source of truth.
- Compress and stream for the target platform; do not let “8K trend” set VRAM.
- Review materials under the lighting the shot will actually use.
That list will still be valid when the next marketing cycle renames PBR. More practical notes live on the blog.
Frequently asked questions
Is PBR being replaced in 2025?
No. Engines still expect metalness/roughness or specular/gloss. New work is in authoring (AI, scanning), streaming (virtual textures), and lighting (RT, GI). The map set stays. Learn it once; apply it everywhere.
Should I archive 8K for every material in 2025?
No. Virtual textures and Nanite-style pipelines can stream high res, but most props still look correct at 2K. Keep 4K–8K for hero close-ups and film. Disk is cheap; VRAM and art time are not.
Are free CC0 libraries still relevant?
Yes. Teams still block with public scans, then customize. Locker3d remains a free CC0 shelf you can legally modify, which matters more as AI outputs need a clean, licensed starting point.