The pros, cons, and techniques of procedural asset creation (renamed)

[Laa-Yosh]

Let's just say there are reasons for jobs like set decorator, home/interior designer and such; and even clutter requires a lot of work to become realistic. And all of these are required if you want to set a mood with a set. No computer can do this, at least not yet.

Sure, a massive RPG might not have need for such things but most FPS/TPS games will focus efforts on the player's path.

As for what failed with us, simply procedurally altering colors in a texture is never going to work well enough, especially if you see a lot of the same stuff near each other. We've tried it with soldiers and it always looked either too much, or was completely unnoticeable. Manually creating texture variations has always been the only approach that worked for us.

 

[Laa-Yosh]

Dont know if it is the case but often repair and tweak takes more time than doing from scratch in so many things....

Exactly...

 

[homerdog]

Really? So having a computer generate the basic layout of a living room and the artist moving stuff around a bit would take more time than making the whole room from scratch? Honest question I really know nothing about this.

 

[pc999]

Really? So having a computer generate the basic layout of a living room and the artist moving stuff around a bit would take more time than making the whole room from scratch? Honest question I really know nothing about this.

In the audio equivalents, many times, yes.

Sure there is things like chorders or Arps, but you already did all the hard work. There is also generative music SW or even just for the arrangement you still wont hear that thing for long. And god save me from looking at that thing note by note.

But nothing is better and faster than do it yourself and well enough to not need to be corrected.

I would be quite amazed if that is different when composing/creating in any other multimedia field (assuming good results are the goal not just semi-passable ones).


BTW I doubt the process happens in the way it seems you think it happens, but I let that to another one to tell US.

 

[Shifty Geezer]

Dont know if it is the case but often repair and tweak takes more time than doing from scratch in so many things....

It depends. Artists already use lots of procedural creation for things like noise and textures, or even terrain generation. Or even, artists take a base model and then change it to generate a new model, such as a new pose. If the generation algorithm sucks and it places chairs on tables and coffee cups in the light fixtures, the cost outweighs the benefits. But get a good algorithm and the player should be able to walk around a town with it looking authentic and without that needing an artists hand for every feature.

Let's just say there are reasons for jobs like set decorator, home/interior designer and such; and even clutter requires a lot of work to become realistic.

Most homes etc. don't use interior designers!

And all of these are required if you want to set a mood with a set. No computer can do this, at least not yet.

It wouldn't have to be perfect. It'd just need to be a step up from the constant repetition of the current systems. A street shouldn't be a street made out of building models 3, 4, and 7 of ten, but of buildings that look different enough to add variety.

Sure, a massive RPG might not have need for such things but most FPS/TPS games will focus efforts on the player's path.

Depends on the game. As with all games, where not every tech is suitable for every game. A linear path may well benefit from a full artist's approach. Anything that doesn't need that level of attention, where the scenery isn't a focus, can get away with the cheapness of automatic generation IMO.

As for what failed with us, simply procedurally altering colors in a texture is never going to work well enough, especially if you see a lot of the same stuff near each other.

I agree. It'd need a better starting point than this-gen-with-colour-tweaks. It's amazing how sensitive people can be to repeating patterns, which is a personal pet-hate I have with some Speedtree. I don't see the fact it hasn't worked well up til now as proving it'll never work though. If we think of NPC behaviour, the holy grail is sophisticated AI, and not designs hand-tweaking every NPC life-cycle and interaction. By that same token, if these AI NPCs could go shopping and decorate their own interiors, the insides of buildings would be algorithmically generated. If AI would work for that in an open world, then it'd also work for on-the-fly decoration where the same AI decorating principles are applied to a room.

If we're to have more realistic worlds next gen, and the generations beyond, we're either going to need a gazillion artists creation thousands of individual homes at exorbitant cost, or a very clever algorithm that does for homes what Terragen did for landscapes. Or we just don't have any improvement!

 

[pc999]

It depends. Artists already use lots of procedural creation for things like noise and textures, or even terrain generation. Or even, artists take a base model and then change it to generate a new model, such as a new pose. If the generation algorithm sucks and it places chairs on tables and coffee cups in the light fixtures, the cost outweighs the benefits. But get a good algorithm and the player should be able to walk around a town with it looking authentic and without that needing an artists hand for every feature.

There is some great generative algorithms for music, it will be "perfect", it is actually a heavy researched field as far as I know... Can you tell me the last time such a thing was used? In any thing? Probably not, it is perfect by the rules you set (eg classical) but it will not bring anything to the game but boredom and in games there is already some "contextual music and fxs".

Dont ask me why, that music is perfect, it isnt even repetitive (if you dont want to), if you hear you will probably be amazed by how it sounds, you still will not save it to hear again.

It is just uninteresting and out of context, it lacks the human being.

Most homes etc. don't use interior designers!
It wouldn't have to be perfect. It'd just need to be a step up from the constant repetition of the current systems. A street shouldn't be a street made out of building models 3, 4, and 7 of ten, but of buildings that look different enough to add variety.

Every home is made by a human.

Anyway I agree that only a few games need that much attention to detail, like only a few home/spaces need a interior designer, all the others can have less detail, artist time in this case, so we can have open big worlds

Depends on the game. As with all games, where not every tech is suitable for every game. A linear path may well benefit from a full artist's approach. Anything that doesn't need that level of attention, where the scenery isn't a focus, can get away with the cheapness of automatic generation IMO.

In music I wouldn't go that way unless I wanted a very specif goal (reactive sci fi fx/music), from what I have seen in other stuff I wouldn't want a automatic/random generated stuff in anywhere.

The very few things I can imagine it being somewhat useful is some sort of "genetic algorithm" for natural stuff like plants trying to replicate how they grow, in a previous man made scenario.

 

[Shifty Geezer]

Music's very different to background scenery. One is a language, and the other is varied visual scenery. They aren't comparable entities. Algorithmic landscape generation is much closer to the idea of algorithmic scenery generation, and that works extremely well. If you look at any housing estate, you'll see the same basic model house tweaked via human behaviour - some have gardens, some drives, some planters or hanging baskets, some messy bins and some very tidy - which should be easy enough to model given a suitable rule-set and adjustable parameters. TBH I'm a bit surprised some people doubt background scenery could be created effectively when you consider how sophisticated mathematical models have become in many fields! And perhaps more importantly, that adding automated variation in games would lead to worse experiences. I'll point a finger at ModNation Racers for how autmoated scenery generation can do a lot of the basic scenery work to add interest without needing a lot fo effort on the part of the level designer.

 

[pc999]

Getting out of philosophy of music...

TBH I'm a bit surprised some people doubt background scenery could be created effectively when you consider how sophisticated mathematical models have become in many fields! And perhaps more importantly, that adding automated variation in games would lead to worse experiences.

I didnt said it couldnt be done, just said that it (at the very least at this moment) it dont look good enough to replace a artist and repair and tweaks is even worst

I'll point a finger at ModNation Racers for how autmoated scenery generation can do a lot of the basic scenery work to add interest without needing a lot fo effort on the part of the level designer.

Exactly what you get is a game with looks of a amateur job, but you are fine with that because it is a amateur job on a game that is to share amateur jobs.


When you look for a more pro game you want a pro look, scenario, level design...


If any, the big revolution in visuals and games will not come from the engines or HW, but from the tools artist have IMO, let it be to do new levels/enemies/animation/basic AI/....Rage is a good showing of it IMO.


I am the only one that miss lot of variety in stuff like we had in SNES games, different kinds of levels, with different enemies with different abilities, and new traps and a touch of personality.

 

[patsu]

I am with Shifty. I think it depends on the game. I don't think realism or exquisite visual quality is the mantra for nextgen; with the casual crowd playing games big time, I'd say the visuals will be more varied.

For a game like Modnation Racer, I would be more curious about a driving game that auto generates a track from Google Map's A-B driving directions. The game may have to create an art style from the photo shots (LBP's cardboard look, or 2D-to-3D image conversion, blah). After all, people complain about Gran Turismo's 2D scenery but we still have tons of fun fooling around with it.

For AI, I would expect someone to extend Demon's Souls. Basically, observe/record the white ghosts (i.e., actual players in their own worlds) in different spots and play them against the current player. This is essentially like the World 3-2 boss where the game summon another player to fight on its behalf (brilliant !). the game can also choose failed moves from real players to add unexpected behavior to the "AI". Naturally, it can steal the moves from low level and advanced users to match the skill level of the current players.

EDIT: Come to think of it, both can be done now. Map auto-gen can be done asynchronously. GT5 supports autonomous racing anyway so the users are already used to leaving their game running for an extended time.

Player recording and remote playback is already implemented today in Demon's Souls using PSN's realtime player info infrastracture. They need to vary the moves, split and organize them. Uncharted's animation blending system may be a good starting point too.

 

[Laa-Yosh]

The start of the discussion was - why so linear games. One of the reasons is that producing high quality content is very expensive, so when a studio chooses that way, they're also motivated to make sure that every player will see it all in the first playthrough. Otherwise a lot of the art budget is badly allocated and thus it's wasted.

All this talk about how and what should be automated or what can't be is interesting, but ultimately irrelevant. Just because you might have lifeless computer-furnished rooms it still won't make a labyrinth out of a COD level.

 

[patsu]

May be balance high quality games with lower visual quality but "more fun/experimental" games ? That should allow game developers to explore both ends. They may have different price point or even business models.

 

[Shifty Geezer]

All this talk about how and what should be automated or what can't be is interesting, but ultimately irrelevant. Just because you might have lifeless computer-furnished rooms it still won't make a labyrinth out of a COD level.

Automated scenery should stop it being lifeless though. Or at least less lifeless. Looking at my Ps3 games in this draw, Resistance, Borderlands, and Dungeon Siege 3 to a degree, could all be enriched with procedural content. Of my PSN games, Infamous would benefit a great deal, as would Dungeon Hunter Alliance. Uncharted won't benefit because everything there is crafted as part of the linear story, as you say. LBP also needs carefully crafted levels. Horses for courses, but it's a tech that should see investment and progress to get the best out of next gen. I really don't want another open-world game that's populated by bland clones of the same handful of people wearing the same handful of outfits and living in the same handful of buildings. We were even promised procedural content this gen but it never came to anything. Future game engines from Epic and friends will hopefully be looking into this on a serious level in the same way we have libraries for procedural game mechanics (physics), procedural behaviour (AI) and procedural animation (Euphoria).

 

[patsu]

Laa-Yosh, I am more inclined to play linear games. I still haven't completed an open world game yet.

I think besides your artist perspective, from a not so hardcore gamer's perspective, I think linear games can be more focused and easier to have fun. Some games will benefit from a more open gameplay (e.g., Open space FPS + dynamic AI), but not all. I think the more restrictive/linear tower defense game is easier to enjoy than a full blown RTS.

I think Shifty is making a different point though.

EDIT:
Just to clarify... It is possible to make an inferior linear game.

e.g., Gran Pulse in FFXIII is more open and fun than the carefully crafted linear section in the first half of the game. But that is a gameplay and pacing issue. Not art related. In fact, Gran Pulse is pretty bland with random creatures and pre-defined sidequests. It is fun for me because it has more interesting combat, more variety, and rewarding outcome. Around that time, all our characters are equiped with Eidolon and have more powerful spells.

 

[Laa-Yosh]

There's no procedural content because the tools are not up to the quality that's expected.
That's what I've been trying to tell you guys but everyone's more clever and thinks that it's such an easy thing... About time to realize that maybe I'm not just making all these arguments up, that there's some reason behind it

 

[patsu]

What do you want to generate procedurally ?

Is randomly spawned creatures a procedural exercise ?

Is observing player's tactics and stealing them a procedural AI ?

Are we after gameplay or just the atmosphere ? An auto-generated level may be unbalanced, but an auto-beautify level may be ok.

Modnation Racer's autogen tracks and auto-populated tracks have a user-generated content angle. So the draw is more the community than the actual content. e.g., When I play LBP, I half expect to see silly levels. And we do enjoy playing them after lowering our expectation. Some of them are pretty funny.

 

[Shifty Geezer]

There's no procedural content because the tools are not up to the quality that's expected.
That's what I've been trying to tell you guys but everyone's more clever and thinks that it's such an easy thing... About time to realize that maybe I'm not just making all these arguments up, that there's some reason behind it

The fact the tools aren't up to snuff now doesn't mean they could never be created. Nor is the current status quo disproof of the concept or potential. The argument I'm putting forwards is games need the tools to push the content forwards, because it won't be feasible to pay artists to create the variety of assets, or else next-gen is going to be seriously hampered in appearance in some games do to repetitious content. Or do you disagree, and feel next-gen's visuals are going to be paid for by employing yet more artists?

 

[patsu]

What specific procedural systems are we talking about ? I can see balancing requires lotsa human testing and tweaking. So we can't really run away from tweaking even if the original level is generated by a human. One of the key questions is can a tool generate an easy-to-tweak level, or does it generate raw and unmodular "spagehtti" ? Laa Yosh seems to imply it's the latter ?

 

[Laa-Yosh]

Is it really that hard to understand my point? Just because we need high quality procedural content creation tools is not enough to overcome their inherent problems, the greatest of which is that a computer has no intelligence and artistic sense.

 

[patsu]

I am afraid it may be hard to visualize the future unless perhaps we pick a concrete (and simple !) example to start.

 

[(((interference)))]

Is it really that hard to understand my point? Just because we need high quality procedural content creation tools is not enough to overcome their inherent problems, the greatest of which is that a computer has no intelligence and artistic sense.

Couldn't you teach a computer to have some idea of aesthethics? Like by just getting it to randomly furnish rooms and select which ones are satisfactory or not? Or showing it thousands of labelled pictures of rooms etc.