Comment by Avicebron
20 hours ago
I feel exactly the same. There has to be a way to do procedural generation in a way that doesn't feel "sparse". A thought that comes to me as I type this is what if you try to co-evolve the ecosystems so it doesn't feel like it's just a mix of the same base components in different quantities.
I think a procedural generated space opera is a bit difficult to do.
Space operas are closer to fantasy than hard science fiction: they are about extraordinary people and their extraordinary interactions, you aren't supposed to think hard about the physics.
Procedural generation, on the other hand, is all about emergent behavior, which is closer to hard science fiction: tweak the rules and see what happens. You have to build some deep core physics that are robust, make all the other systems interact in sensible ways over a wide range of conditions, and not worry when balance goes completely off the rails in some cases (in fact, that's the fun part).
No man's sky is a space opera. You could imagine it being hard science fiction: digging into the ground could be deadly (as opposed to a reliable shelter), water / atmosphere mechanics could change (as opposed to just... water and air), gravity could have a much greater range, resource concentrations could vary more, etc. You'd need some mechanics updates to avoid making this frustrating, but maybe not so many as you need to keep the space-opera plot moving.
I don't know that it's possible. Once you see 5-10 of "new thing" you sorta understand the base layer + different modifiers. We're pretty good at recognizing patterns. It's cool the first 3 times, but then you catch on and it feels fundamentally the same as all the other things you've seen.
All procedural games feel like this to me at least. I've been playing Diablo / POE recently. Same deal, once you're at the endgame - the goal just becomes bigger numbers. The maps have different themes, the monsters are different, but you sorta just look past that part as you're gathering generated gear with bigger numbers. It's why I typically get about 10% into the endgame and eventually stop playing.
Minecraft is also a decent example, at some point you stop exploring, build a base and focus on the progression rather than seeing what the system generates in the next biome.
I think Minecraft gets it right by using procedural generation to add depth to the existing content rather than making different content.
There's a fixed set of biomes and notable geographical features designed by the devs, so there's no incentive to keep finding unseen ones, but each world gets its own layout and landmarks, so you get to explore the details of that specific generation for days without carrying over your knowledge of the previous one.
In the long term it might get repetitive too, but it does at the same rate as one gets tired of any other game I think.
I agree a lot with your point on Minecraft. The farther you explore and the more you see, the higher the bar feels for what's worth your interest.
When you first start playing the game, a medium sized mountain is exciting. Then the bar raises towards more interesting features (a deep canyon or a village), and then towards really interesting features of land. Finally, you look for obscure biomes (e.g. badlands or ice spikes). By the end, you can go thousands of blocks without seeing something that feels worth exploring.
Luckily Minecraft has other aspects beyond just the exploration.
I am but an armchair expert here, but I do think most of what's missing is the complexity of real-life ecosystems and "connections". Having biomes that are substantially large on a given planet and diversity among them, animals unique to each, and some kind of connections between the sparse settlements (even just like 3-5 actual cities per planet would be a start) would be a big difference. It would be a bit less about the 'wonder' and more about understanding the interplay.
For now, all I can see is "optimization" and "system limits on numbers of unique objects for things to work at 60+ FPS".
> even just like 3-5 actual cities per planet would be a start
This often takes me out of immersion in reading sci fi books. Every planet has one culture, one ruler, one major city. I know it's usually just a convenience but has the author never seen earth? Why would future human colonized planets not have hundreds of thousands of cities and competing factions?
Finding only one village on each planet is kind of laughable in the same way. The stage of human evolution where there was one village is such a tiny slice of time (if it ever even existed at all).
> I feel exactly the same. There has to be a way to do procedural generation in a way that doesn't feel "sparse"
A good start is not to try and build an entire galaxy. Procgen that focuses on the fine detail of a relatively small world have a big leg up over these "everything generators" (i.e. think of something more on the scale of Tiny Glade or Bad North)
The best proc gen experience I have had is with minecraft modpack called hexxit. The world is procedural generated, but then there is a library of ~200+ hand built structures/features that are scattered throughout the generated terrain. So you wander through a forest and find a cave into a volcano full of diamongs but guarded by angry dwarves, then head over into the desert and find a castle with a witch king, etc etc
There are enough hand built things that you don't quickly start running into repeats, and they are just melded into the generated terrain.
This reminds me of a mod for Rimworld that adds in other people's failed colonies and decays the structure making a real ruins from other people's playthroughs of the game. Awesome concept
>This reminds me of a mod for Rimworld that adds in other people's failed colonies and decays the structure making a real ruins from other people's playthroughs of the game. Awesome concept
This sounds like it would be a great mod for Factorio. Adds other players factories but in a dilapidated and decayed state.
Reminds me of bones files in nethack.
2 replies →
I love this, because decay implies a story, evidence of somebody else's specific failure, which procgen can't fake. It's not just clever, it's brilliant!
Minecraft's procgen villages and mines are oatmeal by comparison. Oxygen Not Included generates ruined stuff, but it isn't from other players, and I've never found it interesting -- just a way of handing you some advanced objects you don't have yet. Ruins with no provenance are a loot table with extra steps.
NetHack bones files, The Sims Exchange in 2000, Spore's pollination, Drivatars, and this mod are all the same mechanism, reinvented every few years by people who don't know it has a history or a name.
It has a name. Maxis called it the Pollinated Content System, and Caryl Shaw's job title was Pollinated Content Producer. GDC 2006, "Building Community Around Pollinated Content in Spore":
https://www.gdcvault.com/play/1013402/Building-Community-Aro...
"Massively single player" names the shape of the game. "Pollination" names the mechanism, and it's the better word, because pollen goes where it's needed, carried by something that isn't the plant.
Chaim Gingold wrote: "Why not build creative exchange into the game, rather than as a website that orbits it?" The Sims made you visit a website. Spore made it invisible and automatic. That difference is most of the value, and it's the part every mod-based version of this still hasn't got.
http://www.levitylab.com/blog/2011/02/brief-history-of-spore...
Placement is the hard part, and it's the part nobody proposing these mods has thought about yet. Spore's server pulled content in as needed to keep the ecosystem balanced, and the game built a model of your taste to decide what to ask for. Will, in the February 2007 PopSci interview:
>We're basically building a model of the player's aesthetic. So if you keep picking things from the shopping catalog that are like super-cute, Disney-like, then it will start offering you and downloading more stuff that matches your aesthetic... And when it makes a request to the server it's taking that into account, so the world will slowly start to represent your aesthetic choices.
https://www.popsci.com/entertainment-gaming/article/2007-02/...
Caryl Shaw, whose actual job this was:
>I have a team that works on the server like the database schema who figure out the algorithms to do the matching so that the content that you get make sense for your game and you don't have to download hundreds and thousands of mega bytes of content. You only get the kind of stuff that makes sense for your game.
https://www.animationxpress.com/interviews/exclusive-intervi...
That one mechanism answers Avicebron's "co-evolve the ecosystems", ratelimitsteve's "geography never affects the animals", and analog8374's curation problem, all at once.
Curation comes free. Download counts rank quality better than a designer can, because thousands of people are each picking for their own taste instead of one person guessing at everyone's. The "tedious lot of work" analog8374 describes is work the community does unpaid, and does better!
Payload size is why any of this is possible. Spore creatures compressed to about 3KB; 1KB of source expanded to 5MB of content, a 5000:1 ratio. Better than that: every piece of pollinated content was a "card" -- a 30KB PNG with the game data embedded inside the image file. You could email a creature. You could post one on a forum and somebody could drag it straight into their game. The picture was the data. A decayed Factorio base or Minecraft settlement is a small diff, not a 30GB folder.
https://www.gamedeveloper.com/game-platforms/gdc-i-spore-i-p...
Why it keeps dying: the mechanism is cheap to build and expensive to maintain, and it generates no recurring revenue, so it's always first on the chopping block.
Drivatars rotted. The Sims Exchange is gone (but I'm working on solving that problem). Spore's servers are somehow still up, which I always assumed was because there's nobody left at EA who knows how to log in and turn them off.
Now that EA belongs to the Saudi Public Investment Fund (93.4%) and Jared Kushner, the failure mode I worry about isn't a forgotten password.
The PIF's chairman is Mohammed bin Salman. In 2018 his people met the Washington Post columnist Jamal Khashoggi at the Saudi consulate in Istanbul with a bone saw and a forensic pathologist who had explained the plan on tape twelve minutes before Khashoggi walked in: "After I dismember it, you will wrap the parts into plastic bags, put them in suitcases and take them out." He left the building in five suitcases and his remains have never been found. Three days later they sent the consulate car to a car wash. The Saudi government's account changed five times in three weeks; the New York Times ran it under the headline "Khashoggi's Death Is Explained by the Saudis in Five Acts (and Counting)". In 2021 the US declassified its assessment that bin Salman approved the operation. Every word of that is one search away, and reality is worse than any joke I can make about it.
So the thing I actually worry about is not Spore. Nobody is going to convene a meeting about Spore. It's The Sims. Saudi Arabia's media regulator banned Final Fantasy XVI from the kingdom in 2023 over Square Enix's "complete refusal to modify the content to suit the region", and made Roblox switch off voice and text chat and delete over 300,000 games. The Sims 4 has shipped gender-independent clothing since 2016, sexual orientation as a character setting since July 2022, and since patch 140 in January 2023, chest binders under Tops > Tanks and top surgery scars under Body > Body Scars. Go look, it's in the base game. A regime that could not tolerate one man writing newspaper columns is not going to tolerate two little cartoon women getting married in a suburban house you built for them. That content is a rounding error in EA's revenue and an existential affront to the people who now sign EA's checks, cutting it costs nothing and buys goodwill in Riyadh, and the $700 million a year in cost reductions gives them a reason to say it was about money. I am writing this down with dates and menu paths so that when it happens, nobody gets to say it wasn't foreseeable.
But I digress...
"The knowledge is lost in the depths of the netcode" is the symptom. Nobody being paid to own it is the cause.
And generative models are trained to produce the conditional average, so pointing one at the oatmeal problem gets you oatmeal faster and in greater volume. The only fix anyone has ever demonstrated is a person who cares about one specific creature.
Kate Compton says 10,000 bowls of oatmeal. Phil Hartman on SNL got there in 1989, and it was a lot more than 30,000.
Stop settling for oatmeal. There's fiber, and then there's high fiber!
https://www.youtube.com/watch?v=Ku42Iszh9KM
I don't know why but it made me think of the original premise of Drivatars in Forza games - Microsoft selling it as an "AI" learning how you drive, then your individual driveatar would be submitted to Xbox Live and would randomly appear for other players in their offline games, but it would drive just like you drive. Really cool concept, of that "footprint" of your play appearing in other people's games.
Shame that it just languished to a point where I think Microsoft completely lost both interest and control over that feature, where in Horizon 6 it just gives you the exact same few driveatars every race, and somehow no one at Playground Games is interested in fixing this. At this point I strongly suspect that they literally don't know how to, the knowledge is lost in the depths of the netcode.
That sounds a lot like that Dungeon Crawl Stone Soup approach of randomly generated floors/levels with designed "vaults" emplaced within them
I actually created a Minecraft server plugin a long time ago which did this. It allowed server owners to create their own dungeons/buildings/etc and they would generate randomly in the world. It even supported custom procedural generated dungeons and structures.
Mine was a procedurally generated FPS I've forgotten the name of, it used a seed key to generate the map and then you could fight on it, if it was a cool map you could just keep or share the key so was replayable.
Iirc it used a voxel engine so would have been late 90's.
I play Tekxit for the same reason. The world feels genuinely alive and surprising. I just love automation as well, and Tekxit hits that sweet spot of "building complex resources extraction and refinement chains" but also "a big world full of adventure where you can actually use those resources for something other than making bigger machines to implement bigger processes to build more complex extraction and refinement chains".
I think it has to be hand crafted with some mix of procedural generation. I worked on the Division Underground DLC, and the way we did it is we had artists create dozens of various room variations, describing how they could connect each other, then I wrote an algorithm to populate the map with these "dungeons" and their rooms in a semi-random order. In the end I think the map had about ~200 unique dungeons, but rooms could repeat between them, so you'd sometimes go "oh hang on I've seen this metro station before" but the entire dungeon would be different almost every time, unless you played hundreds of hours of it(the game kept track and wouldn't give you the same dungeon again until you played them all, iirc)
I think you're on to something here. The problem with procgen is hard because the various things in an environment have to interface with one another and in NMS they don't really meaningfully interface. The geography is generated, and the animals are generated, and the mineral deposits are generated, but the geography never affects the animals or the mineral deposits. You don't see, as an example, vast highland regions with animals adapted for living in the cold and huge mineral veins exposed in cliffs at high altitudes. You also don't see older mountains where the average mineral deposit would sit lower in altitude but be smaller due to erosion, with more cave systems and more critters evolved specifically for caves (colorless, blind, chemautotrophic, specializations like that). There's a range of ways the landscape can be, there's a range of ways animals can be, there's a range of ways resource distribution can be, but they don't affect one another and it makes it feel bland and same-y.
I feel like this is where AI would come in handy. An AI can get a lot more 'creative' than a procedural generation algorithm.