The q3map2 “fast” lightmap falloff produces very ugly lightmaps but we cannot assume everyone can spend hours or days on map compilation by not using “nofast” lightmap falloff.
For example my own old computer compiles the vega map in minutes with “fast” lightmap falloff, but compiles in two days without it. Even if a ThreadRipper 3990X may reduce it to 15 minutes given some prognostics, we cannot ask contributors to spend 4000$ in a CPU (and another 1000$ for the things around).
So I driven some tests to try to approach the “nofast” looking while keeping a “fast” like build time.
After hours of compilation… the answer is fairly simple: we must build the first light step without “fast”, but build all bounces with it. Technically it means to not use the -fast switch but to still use the -fastbounce one. Those are both options for the same q3map2 light stage.
This produces almost the same good looking of a nofast build for the cost of a couple of minute on an old laptop and some seconds on a beefy modern workstation, literally saving hours (if not days) of compilation.
See for yourself with the chasm map:
