Lander wrote: ↑Fri Jan 19, 2024 3:30 pmWell, I suppose I am something of a selfmade proc-gen bigot. My surface feeling is that the further you stray from human intentionality (which frequently implies bolting together samey modules Risk of Rain style,) the less chance of good remains.
The core issue I think is most people approach it as a way to do less work. "Eh, instead of designing levels I'll have the computer do it." As laziness is the #1 virtue of programmers (double-edged knife that it is, that requires an over-abundance of the 2nd virtue, impatience, to counteract its downsides), I get it.
Simple algorithms made sense when program sizes were limited. The "nine boxes connected by corridors" used by Hack, Mystery Dungeon, Lufia 3, and probably many others are pretty terrible. Less efficient than just having every room the same size like in the first zelda with none of those one tile wide corridors. But the simple algorithm used in Pitfall! was pretty alright. It was a way to make a game bigger than you otherwise could have.
But the core truth is to have ~ten times the variety, you have to do ten times the work. And you have to identify what matters, and what does not matter.
Let's take a castle generation algorithm: every room in a castle should have a purpose. Entrance hall, audience chamber, barracks for soldiers, hallways, toilet rooms, guest rooms, kitchen, etc. They should be connected at certain places, and some should only show up a single time in a single dungeon. Whether an entrance lobby is 29x30 tiles big or 28x31 doesn't matter whatsoever. But the decorations inside matter quite a lot: if there are portraits of bunny rabbits or suits of armor tells a lot about the personality of the castle's owner, and can be a tool to create that sense of
anticipation thing. Those statue heads in Zelda's dungeons do a lot of work.
Platformers are probably the hardest, since you have to be aware of many meta-characteristics of a stage. One basic example is the Mario tutorial obstacles, where there's an easy and safe version of a challenge. Followed by a harder version, and finished with the hardest version that probably has some of those deadly Pits added to it. That kind of thing has to be intentionally put into the algorithm, or it'll never happen.
It's a bit of a trip to think the design of the houses of the NPC's in Animal Crossing are better quality than these empty hallways with monsters and treasure chests.
Randomized terrain has been a staple of AAA games for decades now - it doesn't matter so it saves them a ton of money on generating empty space. There's even that random tree generator that was industry standard last time I checked decades ago. (Gamefreak could probably use a tool like that. To get some consistency in their tree art.)
The box model gives a massive disincentive in making games replayable like this, though. Most people will only play halfway through at best, and then you want them to buy the next box. So all this is really only relevant to indie development.
As for human intentionality, consider the games that are basically indistinguishable from something a lazy half-assed random generator would spit out. At the end of the day it's just another tool a creator can use, and if they don't know how to use it, it's worse than not having it.
It's more like gradients than dithering. You can never have enough dithering, but you can make some vicious eyesores with too many gradients. (Some Atari 2600 homebrew guys sure do love their gradients, man.)