Let’s start with introduction:
My name is Alex and I’m working on my unannounced first-person indie game as a solo developer. The game is based on a real-life location and must have several unique real-life landmarks.
The problem begins with converting large scanned 3d models into optimized game-ready assets while maintaining the uniqueness of the landmarks.
Things I already know (please, correct me if I’m wrong):
- 1 draw call per material.
- 2k is the standard resolution for textures for FHD gaming. 4k is overkill as only few people can play in native 4k without any upscaling technology.
- 10.24 px/cm is the standard for first-person games. A drastically lower number will result in blurry textures.
- UDIMs are bad for realtime performance. It’s mainly used in VFX/Cinematics.
- Asset modularity is key for optimization.
- Tileable textures and trimsheets are also key for optimization.
- You can use several UV sets. One for tileable textures with correct texel density and one for grunge/dirt masks inside the engine for additional details.
- Decals are also a great way to add extra detail, but since they use additional materials, they also increase draw calls.
Things I’m trying to understand:
I will provide several screenshots to try and communicate my problem clearly.
This is scanned landmark I’m currently trying to convert into game-ready 3d asset. My goal is to keep as much of the stone structure placement and window/door ornaments as possible.
When I started this project, I thought I would simply create low poly version of the building and bake all the necessary information using substance painter. But as I learned there were things such as optimization and texel density, my workflow started to fall apart.
When I heard about UDIMs, I believed it was the correct approach, but reality struck again with bottlenecks such as performance and memory optimization. (You simply cannot have 100s of 4k textures and expect to have 60FPS)
After fiddling around, I tried to divide the model into several modules(pillar, wall, wall with window, roof, etc.) but couldn’t quite get how to connect front and back with other pieces and fit them on the grid. Plus, how to keep 10.24 texel density on 4-meter-tall building while not using tileable textures (tried to create trimsheet shown on the screenshot, but it only tiles horizontally and doesn’t really do much with my modular pieces)
My previous experience:
I have had such dilemma in the past, but it was with hand modeled asset.
Solution with that example for me was to use tileable textures and several materials (all the textures were created from 3d scans and the asset was modeled with respect to the reference, that’s how I kept the familiarity with original “landmark". I still don’t know if it was the right call)
here are some of the screenshots:
Conclusion and question:
At the moment, I think the idea to keep as much original information from landmarks as possible for game environment is dumb and not practical. I should use modular pieces and tileable textures which in return will not represent the building truthfully, but that’s the sacrifice I have to make.
I guess my question is how more experienced artist would approach such task, when goal is to respect the reference and maintain as many features from original as possible while keeping optimization in mind?