One man's blunderings into the 3D realm
28 Mar 2020 -
It’s been a long time since I made a post and after looking out the window it might be a while before normal life resumes, so why not write about a subject that’s been on my mind for over a decade?
I’ve been interested in procedural generation since before I learned how to code. My first project with Python I tried to generate random maps out of prefab tiles with Pygame. I failed miserably, but the the concept behind it has stuck with me and I’ve always end up coming back to it. During the years I spent working in the video game industry I produced a number of tools for automated content generation but these always took input from an artist - they never generated something from absolutely nothing. I can understand the urge to make a “magic box” where nothing but perfectly curated content comes from.
That being said, I find myself getting frustrated with how procedural content generation is typically used in the game industry these days. There seems to be a tendency to opt for breadth instead of depth - I would hazard to guess this is done for a few reasons:
If you’ve taken the time to develop a procedural content generation tool then it makes sense you’ll want to leverage it as much as possible. I can also appreciate where the “more is better” psychology comes from; seeing tens of pretty assets generated before your very eyes is pretty amazing. Why stop there when you can make thousands?
Along with the above comes with bragging rights. Big numbers - or even mind-bogglingly huge numbers - are kinda impressive and something to talk about. They can form a nice “back of the box” story to assist with the marketing of a title. One can look no further than No Man’s Sky and its 18 quintillion planets; the procedural generation was part of the core marketing campaign.
I suppose if you were to make an engine that can produce an infinite amount of something and used that to market the game - but didn’t actually give the consumer this magical box you hyped, they’d quietly get the shits. That makes sense - you paid for the game, why not show them the magnificence of the generator in its entirety?
Personally I find this method of procedural generation, ironically, limiting.
Other writers have done a better job at explaining why this is but I find Kate
Compton summarised it succinctly with her oatmeal analogy.
It highlights that even though we might have a fantastic amount of unique and
varied content this doesn’t necessarily mean what we’ve produced is particularly
interesting or deep. Funnily enough a common complaint from people reviewing
such a system is that once they’ve seem one asset, you’ve seen them all. The
generative algorithm and its limits become obvious to the player and they can
almost start to see what the system may throw at them in the next iteration. If
the grass is green on this planet and red on the next planet, it comes as little
surprise that it will be a different colour again on another planet.
Another issue is that techniques for generating art are much further ahead than techniques for generating story or game mechanics, or developers choose to invest in art generation exclusively. This leaves most developers with a glut of art, but because there’s no meaningful story or gameplay to reference it the experience seems hollow. For example, let’s imagine that our generative system creates a tower in an empty field. The system may identify this as a place of interest and put some sort of reward at the top like a unique item or some in-game currency, or alternatively, nothing. Even in the best case scenario the reward feels empty. The tower stands in the world as it does in the field, unreferenced and unnoteworthy; npcs don’t reference it, no story includes it. Its existence in the game world is shallow and perfunctory. Once the player finds the tower and climbs it, the tower’s story is over.
Some attempts of infinite content generation work better than others, typically when the interplay of content gels well with other systems. Invisible Inc feels fresh and interesting with each map iteration because the layout of the space is intrinsically linked to the movement mechanics. Random dungeon layouts in Diablo are less interesting as combat generally proceeds in a similar fashion regardless of dungeon layout. The much touted Spelunky uses what some consider the gold standard of procedural generation, a cleaver mix of generated and hand-crafted content (and one of the main inspirations for my own project).
I’ve started working on a new procedural toolset that I hope to write about over the coming weeks. I’m not sure if I can create a system that produces interesting and deep content but there’s definitely some ideas I want to explore.