Thread by @MOOMANiBE

Okay so I'm gonna do a brief thread on procedural generation in games from a dev PoV. I've wanted to do this for awhile but the concept ballooned infeasibly so I'm just gonna do it completely off-the-cuff instead and we'll see how things turn out.

So a lot of times you see procgen talked about in games, it's talked about as if it's a feature. Oh boy! This game has procgen! That'll make people enjoy it!

Well, no. It's okay for marketing, but from a design standpoint, procgen is an enabler. It's used to make things.

And this comes to the heart of the issue, i think. When you're adding a feature to a game, the first question you should be asking is "how does this make my game better?"

Procgen is good at.. 1.5 things.

1. Making A LOT of something.
1.5. Maybe that thing is unexpected? Maybe?

Let's drill into 1 first, because that's the more practical of the two.

When do you need a _LOT_ of something in games? Like, a LOT. A REALLY HUGE AMOUNT.

A: When each individual player is going to see that thing a TON OF TIMES.

Let's use an example here. Let's say you're making a lengthy metroidvania and you want the entire world to be generated randomly when you start a new file.

How many times does each player see that procen?

Statistically? 95% see it once, ever.

At that point, who's getting value out of your generation?

1. People willing to play an entire metroidvania through multiple times.
2. People who talk to friends who are playing the same game and find differences novel. Maybe??

How much effort did you put in for this impact?

The best value you can get out of procgen is when players see that generation again and again and again, visibly.

Roguelikes, classical ones, are great for this because the permadeath structure means that players will see the start of the game hundreds, thousands of times.

But here's the rub.

A lot of times, you can just _make a whole lot of a thing by hand_ way, way faster than you can teach a computer to make that thing procedurally and NOT be garbage at it.

Procgen is hard. You're not just developing a game, you're developing a game that then _MAKES ITSELF_.

If you want good level design, you need to be good at level design, you need to have clear level design goals, and then teach a computer to make level design that good.

THAT'S INCREDIBLY HARD. YEARS AND YEARS OF DEV HARD. Doubly so if you yourself are _not already good at the skill in question_.

One of the biggest mistakes I see devs making is assuming that procgen can fill in their own gaps in skill. It cannot. It can only exacerbate them.

A lot of the highest-profile examples of great procgen are incredibly restrained with their generation.

Rogue and Angband pull from restrictive enemy/item tables and have simplistic terrain.

Spelunky hand-made all the area tiles and combines them in highly-specific ways.

That's because they understand that making a platformer or an RPG where _everything_ is randomized is a herculean task. Better to make interesting design yourself and then randomize it JUST enough that it feels fresh to players.

That's all the procgen you need.

....This is basically a postscript, because I think I've found my point, but wrt point 1.5, "but it generates novel stuff!"

Novel stuff is novel in all senses of the word - specifically the "unusual" part.

To make a game with a procgen engine that creates surprising things, you have to be willing for the generation to be a pile of garbage 10 times for every 1 that it's cool - at best.

This works only if you create an environment where players are willing to comb through that chaff.

This is basically how dwarf fortress works.

DF is, in many ways, a novelty-creation engine. That's great and fun in isolation, but it's also why players bounce off it fairly quickly.

How many ppl are willing to read through hundreds of history logs to stitch a story together out of mess more than once?

I love DF, don't get me wrong, but it takes a specific subset of ppl to do that, and you have to deeply be clear about what ppl are getting into.

Not a lot of games can manage that alongside more traditional structures. Would DF have succeeded if it wasn't free for decades?

Anyway, consider that before betting on novelty-creation as part of your procgen plan.

Alright I'm done, I think I've said all the things. Thank you for attending this ramble.