And perhaps you are severley underestimating the effort large-scale world assets take to create.

I agree that doing procedural generation properly takes a lot of work, but there's a reason that AAA open-world games like GTAV cost hunderends of millions of dollars to develop.

There's a reason that AAA open-world games like GTAV are using rather limited procedural generation, too.

The problem is that to create N pieces of satisfyingly unique procedurally generated content, you need M pieces of content that were carefully designed within additional restrictions to make them combineable in various fashions - making them harder to create. Then you add code to glue that all together. If you think M=0, that just means you've hardcoded your content in code.

When N >> M - e.g. you're generating thousands of similar-yet-distinct roguelike levels, trees, etc. - procedural generation can pay off.

But even for large-scale world assets, it's frequently the case that N < M. Making the content non-procedurally is going to be cheaper - fewer limitations to worry about, and fewer pieces of content to actually generate. A few clever tricks - rotating, retexturing, tweaking a base model, etc. - let you make a lot of non-procedural content without creating each piece from scatch, and lets you stretch that content to appear like you have more variations than you actually do.

I don't think No Man's Sky will show a significantly better RoI, in the long run. Star Citizen is certainly racking up the bills.

We could even argue that AAA game studios using hand-modeled art assets instead of procedurally generated content is proof that hand-modelling is the "cheaper" route. Though most AAA game studios are using a hybrid approach and have some form of procedurally generated content in their pipeline, whether it's to create quests or to create the baseline environment on top of which bespoke content is added.

Projects are hard. Full stop.

But it does scale better, doesn't it? Seems that would be the case if more tools and libraries like this one were open-sourced.

Not really. The problem with procedurally generated content is making interesting iterations. Sure, you can have a million space ships just as easy as one, but will anyone care about them? It is a different sort of work, but of equal effort, to "just" modeling the content.

Even most landscapes on Earth (reality is "procedurally generated" by the laws of physics) are pretty boring. Spectacular documentaries must travel far to reach the interesting ones.

Interesting procedural generation may require strong AI, because it must model how it is interpreted by us.

> Even most landscapes on Earth (reality is "procedurally generated" by the laws of physics) are pretty boring.

Strongly disagree! I truly feel sorry for you if you can't see beauty and spectacle everywhere you look in nature.

Well, for now, yes- but things like this tend to get easier as new techniques are perfected and more powerful tools are developed, whereas building a thousand art assets by hand will remain time-consuming indefinitely.