Goodbye Handcrafted Levels

Apart from a tendency to use jargon that I find a bit ugly, I thought this was a great essay on game design. The super-short version: handcrafting locks down your gameplay too early in the creative process, and the result are games which try to sustain your interest with a stream of “content”. Procedurally-generated content, OTOH, requires you to develop your mechanics so that the gameplay will work regardless of the combinations that the random part of the system throws your way.

This is why, IMO, Stone Soup has so much more appeal than any scripted RPG, Spelunky is so much better than any ordinary platformer, Master of Orion is better than Imperium Galactica, why campaigns in RTSs suck…

Actually, random generation won’t get rid of handcrafted levels. In some games, maybe, but that’s been true for a while. What devs should be doing is using randomly generated levels as a starting point, as much for the reasons you stated as to save a huge amount of work. Hand crafting the base layout? Waste of time. Taking an existing layout and adding soul? That’s what making games is.

Just filtering out some of the busywork. Honestly been looking forward to the procedural revolution. Working on it in my spare time.

On the other hand, procedurally generated content is soulless and repetitive.

Randomly-generated levels have been around for at least thirty years. Why would they supplant hand-crafted levels across the board now?

Apparently it has something to do with baby food.

I find most completely procedural gameplay just as boring as overly-scripted stuff.

My favourite approach is when developers make solid, flexible, rich mechanics, then design levels around them.

Because it’s cost effective. With content creation costs rapidly becoming unsustainable, randomly generated content is the fix. Well, part of a fix. Plus, it actually paves the way for new and interesting gameplay experiences.

But like I said, it doesn’t have to be soulless. Used as a starting point, it gives content creation teams a huge leg up. Every game could start at the size of a game like Oblivion, and then while old style game development spends 3 years creating Oblivion, new style game development can spend 3 years polishing Oblivion.

Also, procedural doesn’t have to mean boring, or even unscripted. It is possible to have scripted procedural systems.

For example, if you have a game whose story requires that the player get ambushed, right now, either a tunnel leads you to the ‘ambush’, or you follow a map marker to the ‘ambush’. Or the game in some other fashion makes sure you land on a specific spot.

Whereas, with proper procedural methodology, you could map all possible ambush locations in a game that meet the parameters set out by the mission designer, and the next time the player enters one of the locations that matches the parameters, bam, ambush. Surprise the player every time. Except, there’s even an added benefit – a tactical player may recognize the signs of a possible ambush because of the environment, and thus prepare in a way that’s not possible with current games.

The reality is that procedural systems have a lot to offer, but very few people are thinking about them very deeply.

I always thought that procedurally generated content with bits and pieces that were manually created (or polished) would allow for large game sandbox gameworlds with a focused story.

Remember how huge Daggerfall was? You had the option to wander for hours in one direction or fast traveling as the mood struck you.

Well, I don’t think they’re about to, and I don’t think the author of that article thinks they’re about to, either. And I agree, it is quite possible to have soulless, dull randomly-generated stuff… but I think there’s a good case to be made that the “playable life” of a well-made game with procedurally generated content has the potential to be much greater than for one with fixed content. Minecraft is a great recent example. What makes it possible for a one-person project to be so gripping over a long period of time for so many people? Good ideas, sure, and some interesting combinatorics or whatever the word is, but part of it is just that he doesn’t need a team of ten people making maps and designing locations. So long as map-generator is done well, you get just as vivid a sense of exploration as you would from a handmade game - or more so.

…and in the meantime, what he’s working on are the mechanics and the design of the pieces of that world so that the kinds of situations that crop up will be interesting & lively. Whereas in your typical handmade environment, what you find is yet another iteration of the waves-of-shootables and monster-closet FPS stuff that’s been done to death for over a decade. Because making content is so expensive that you can’t afford to experiment with the gameplay. And, as he points out, if you get an inspiration 3/4ths of the way through a project, you can’t try it out because it might break all the levels you made already.

It’s worth noting that I largely disagree with the article, though many of his arguments are sound. The idea that static level based games are shallow and procedurally generated content is deep is actually outright laughable. It is rarely so cut and dried.

I went and played his game, and it’s a pretty cool game, to be honest, but it’s not particularly deep. Or rather, the complexity comes from the positioning of the plane and the speeds, and a rather simple group of variables. In this situation, he’s absolutely right that procedural generation is the best scenario – after all, there’s only so much ‘soul’ you can add by hand placing craft in starting positions.

Really though, it’s not one size fits all and it only works for a very small subset of games.

I think there is a lot of untapped potential lately, to eliminate the soulless lifelessness of procedural generation…

Particularly, Google has digitized and made available a vast corpus of out of copyright works. Lift random dialogue for npcs from conversations in novels, use tone analysis to tell whether it’s an angry scene or a happy one. Fill street walls and bathroom stalls with poetry from the ages.

Weight it’s chance of appearance on popularity, to avoid the dreggiest dregs… Back it with a notebook feature that allows you to find out the work of literature a particular level or part of a level is being lifted from, and it’s a way to organically spread and popularize literature.

I also think there is a vast amount of free 3d models available on Google maps… randomly generating interior maps would provide a near infinite amount of base content, allowing coders to focus on complex random generation of gameplay details… or as suggested above, tweaking them.

Just how high are you?

I think I (and I’m sure many others) have said this before, but I would really love to see Spore-like character creator technology in an MMO used for generating NPCs and mobs that don’t all look identical.

It is true thatprocedurally-generated content is almost certainly The Future, but we aren’t there yet.

I’ve yet to play one game with procedurally-generated content that was anywhere near the quality of handcrafted. Every one of them has been some degree of lifeless/soulless (a huge forest to explore with nothing in it but the same tree repeated forever).

The best example of why we aren’t ready is comparing the environments in oblivion to the Nephim total conversion of oblivion. Although comparing oblivion to gothic 2 or even freaking risen also is helpful.

Gaming at its best now requires heavy scripting that isn’t super obvious. Demi god half brothers should ambush you at predetermined points in the game, but those points shouldn’t be so in your face that you feel like exploiting the triggers is part of the game. Maybe in the future we will have super ai programs that meld the story around you on the fly, but until then… got to do it by hand rather than not do it at all.

To a lesser degree, this is the approach I took with the random map generator in the Heroes 3 expansion. The RMG used scripted templates for maps. The templates specified things like zone size, monster difficulty, treasure value, and which zones connected to which zones. In effect the the template was a very abstract map. The precise physical relationships between zones weren’t specified, just the logical connections, so the map generator was free to resize them and relocate them, and provide teleportation gates if it was necessary. The “random” part was the fine detail of the map, decoration, specifics of object placement, monster selection, and treasure selection.

This is not the same as having a scripted story, but you could add scripted story elements to a program like this.

EDIT: And what about the random encounters in Fallout 3? Many of those weren’t in any fixed location.

Procedurally generate what players expect to be natural, hand craft what they expect to follow predictable designs.

I’m not against randomly-generated levels. I love a lot of games that have them. But I also love a lot of games with hand-crafted levels, and I also dislike a lot of games in each category, because neither one is inherently better than the other. Randomly-generated levels are good when they’re in games that they’re the right choice for, and the same is true of hand-crafted levels.

Two of my favorite games from this year are Fallout: New Vegas (a painstakingly hand-crafted world) and Din’s Curse (in which basically every single thing is randomly generated). They’re both awesome games, and neither one would have worked with the other style of level creation.

Yeah, this is a war that was fought a long time ago. With the exception of strategy games and the cult underground of rogue-likes, hand-crafted content mostly won.

It’s not hard to see why. Creating 100% procedural content means reducing each facet of content creation and game design to an algorithm. That’s hard, and time consuming, and risky. Rather than wait for someone to finally work the bugs out of their General Theory of Crate Placement, it’s generally just easier to go in and hand-place the damn crates where they work best.

As for saving money, it doesn’t really, unless you’re talking very small projects like roguelikes. For AAA titles the primary expense for content is not level designers but asset creation – meshes, textures, animations, etc. (Clever level designers actually save you money by figuring out how to re-use existing expensive assets in ways that never occurred to you before.) Procedural content would replace cheap level designers with programmers, who are more expensive. And it’s increasing risk, since it’s easy to calculate how long it will take to hand-place crates, but very difficult to estimate how long it will take to come up with a functioning General Theory of Crate Placement.

I love the idea of procedurally generated content, but the results outside of strategy games so far have been either horribly unbalanced/unpolished (most roguelikes) or so innocuous as to be pointless (Diablo or Torchlight, which have random levels but the randomization doesn’t actually do anything interesting.) Procedural content has its place, either as a starting point (landscape generation) or as a diversion (e.g. a random mission goal to add spice) but I can’t see it replacing hand-crafted content until it becomes substantially better.