Technical Artist Bootcamp: Introduction to Proceduralism

[Music] big ant for Ozzy he's doing a great job and then yeah so yeah the joke is that Jeff finally passed the baton over to us and it's goods great and we're looking forward to kind of the next few years of GDC and so my name is Luis cruel and I'm a senior technical artist of side effects side effects is a software company I and I'll go into that a little bit later but welcome to my talk this is a great crowd I'm super excited to be here so what I'm gonna be talking about a little bit I'm gonna talk about the definition of procedural ISM it's kind of like a weird buzzword right now to where it's kind of like PBR and there's there's people think that it's a magic silver bullet and everything can be solve a procedural ISM then I want to talk a little bit about how we will roll out a procedural 2 system and basically you're going to start out by basically building tools for you and then you're gonna move on and building tools for others and then you're gonna start building tools for everybody and then there's kind of mind shifts that change when you kind of hit each of those

stages my background so I was in game development for about 10 years and then recently I switched to be a software development and my job is exactly the same so basically before I used to write tools and maybe like 10 20 people in the studio you use them and then every now and then I'll get an email be like hey this is broken can you fix this come to my desk same thing happens but now it's just the whole industry which is awesome so I write tools and then the tools get built into Houdini and then everybody can use it and then those throw email me with broken stuff and I'll go to fix them and I'll go visit them and help them so it's it's an awesome it's a really great job and this is not a Houdini sales pitch so I'm gonna try as much as possible to kind of keep it up here to wear it as a procedural software system you could implement this and anything with scripting but just kind of like at a high level just I'm gonna mention a little bit of Houdini but this is not like entirely about it so what the hell is procedural content generation and then some people call it PCG which is procedural content generation there's some talks in the AI summit that I'm going to be immediately after so there's this weird kind of

misconception of like what is this and how does it work if you're familiar with substance which most people here are that's procedural ISM so it's basically you're writing little algorithms to make content and that's the simplest of it like I would even say righto rigging procedural ISM because you're basically just writing procedures to build content at the end of it so this is a couple of examples so Brad Smith as an awesome artist used to be a naughty dog now it's a topic and then he just does a bunch of substance designer stuff and yeah we were building art with these little node networks to kind of make content and then Magnus Larsen does the same thing but instead of using 2d he does it in 3d and that's where Houdini comes and it's the same kind of workflow is just you're you're working with nodes but instead of doing 2d operations you do in 3d operations so this is the only Houdini slide that I'll show just for people that are not familiar what it is the same operations that you would have in any other software like extrusion creating primitives beveling so it's all basically notes that you just drop down

and then you would use it like you would and then the nice thing is that it's non-destructive because of the procedural nature so you can kind of go back up to the level the top level change the input and then everything would just kind of calculate itself and updated so here I just have a simple sphere and then I have a cube that I can kind of boolean them together and move them around and then at any point I can go into that chain and then change the inputs so this is a super super basic tool where you get the idea of if you're kind of familiar with the kind of what substances to photoshop Houdini is - any other kind of 3d software app so at least now you guys have some context of what I'm talking about the other thing too is there's different types of procedural ISM that's where some of the confusion comes from so there's like offline cos procedural ISM which is like the substance and Houdini thing that like basically we're building things offline and there were just baking static meshes that go into a game engine there's low time procedural ISM which is kind of like Minecraft that's like while the game is loading things will get generated and then they'll get baked down into static content and then there's runtime procedural ISM which is kind of like the the vibe and the tilt

brush to where people are sweeping curves and then like as they're moving their hands a procedural is kind of taking that curve input and then doing things to it and placing a ribbon around it for this talk I'm going to focus mainly on the offline because I feel that's the most relevant for kind of game development as a whole sure some people might be already kind of going into the runtime and the low time but for like 90% of the population I think that the trick is just how do you automate your pipelines and how do you make them a little bit so I'm hoping that out of this talk you guys can get at least enough ammunition to where you can go back to your studio and be like hey this is something that we should at least think about and kind of further explore a little bit so why should you clear you like crunch have you ever crunched you will have you ever lost work because someone came back and it was like yeah that the artists final but design had changed that we played that aside in a play like rap so we have to go back and we have to change it and then you just have to eat that work and you just have to rebuild everything and then you have your habitat to cut stuff because you like it just couldn't have enough man cycles to

build something because the idea was too big or there's just way too much work to be able to pull it off so the idea with procedural ISM is that then you try to minimize all of those you're still going to have crunch probably you're still gonna have a little bit less work but the idea is that if you have a kind of automated pipeline that you can change the input in the beginning and then kind of have things fun away the way through ideal you're minimizing all of these problems so the first misconception is that procedure content generation is this weird ball thing that Indies do for roguelikes with procedural dungeon generation sure that's also true but that's not exclusively what procedural content generation is and then on the flip side people say procedural content generation is this thing that you need a whole development team for so like you have the snowdrop engine that you have like volumes that you can push and pull and building generate and then the whole cities get generated automatically so there's this it's they're both true but they're also both false so they're the scope is so big that don't get too hung up on the definition of it just kind of figure out how can I apply this to me don't look at too much of what other people are doing just use it as

inspiration but try to kind of frame it in your own problems that you're trying to solve so the usual evolution that I see in kind of different Studios and it's kind of fun to see where the stages of development of kind of opting procedural ISM is usually you start with one kind of champion that is like I have a crazy idea and I used to say that I like I have this crack-smoking idea that is just like off the wall and I have that I'm gonna pitch this for this whole crazy system and then they'll do it and they'll pull it off and then now people start to notice and they're like holy like that that actually worked and then you're gonna have like a few people kind of gathering around them and being like okay how do you how did you do that can you build this something slightly different can I use your tools to do it and then those people start using and that's gonna be like the badass theme of the studio and that's the hard problem team and they're basically like the the Wizards and then ideally you kind of be able to distill that out into a production at large and that's the really hard part is basically kind of going from this kind of small team agile movement to like a big engine deployment kind of level so I'm going to go through

each of these stages and kind of help you through at least know what you're gonna run into and kind of how to kind of muscle your way through or just understand what the problems are going to be so we'll start with proof of concept so this you're gonna see is a lot like just regular tool development so you just identified the problem figure out how to make time to fix the problem and then you fix the problem so how do you identify the problem walk around and this is like if you guys are tech artists you know just walk around to the tool the studio and figure like talk to people and walk around and then the one thing that I would say is for procedural ISM in particular find out what's boring and repetitive because I'm gonna talk a little bit about this but some people are scared of procedural ism and they think that that's going to displace their jobs but the idea here is you're taking things that no one wants to do they're kind of just shoveling assets through pipelines and then all the boring and repetitive things and all the little bottlenecks are the perfect candidates to be automated because if something is so repetitive that basically the artist can just shut off their brain and then just do all just yeah I click this then I click this and

I click this and then I move this file to here and then I bring this file over here that should just be automated and that's a perfect candidate for you to hit the other thing too is understand the arc flow so people joke that I say that you have to basically talk to a robot or I talk to artists and then build robots that will replace them that's not true but the idea is that you do talk to artists and then you figure out what the workflow is and any try to automate as much as possible while leaving control points and that's the crucial part is that you want to build an automated system but you want to leave control where the artist wants control so you need to kind of expose the hooks to where any point in the pipeline they can go in and they have feel like they have control and then go through the process yourself and this is especially good for tech artists because we can actually do it so sometimes with like tools programmers they'll just get a list of this is all the stuff that we want and then they'll just go check let's check let's check with checklist and just give you the tool and it's like yeah technically that's what I want to do but you missed half the point and I forgot to say that yeah there's this other pain point so if you actually do the process yourself you will see that

yeah no you guys don't even bother doing any of this just like completely Hotwire the whole thing and go from A to B way faster this is it's fun because it is the same process as a tech artist - so a little bit yeah so the misconception is that I'm training my replacement and this is an actual conversation that I had with someone to where I was like okay talk to a designer how do you build procedural levels and then I guy was like listen man I went to China one time and then I had to train a replacement team to basically I came back and then I got laid off because basically I trained my outsourcers so I feel like I'm having the same conversation and this is and this is not just in the games industry just like out everywhere you hear a lot about automation we had conversations last night about it just know that people are gonna have that kind of built-in resentment and built-in resistance sometimes and you just kind of have to know how to word it the way I like to word it is we don't want to take your job we want to make you more productive we want that's why we build tools and then it is the same way but there's a little bit of a mind shift that needs to happen with people with

procedural system they they don't see them necessarily as tools they see them as the next step of tools so there's a little bit of that and then there's this thing as the goldfish paradigm which kind of ties into that which is you will build as much content as you have time for so the goldfish will grow for as much size as the fish tank is so the fact that now we kind of displace the time and we may production shorter that just means you're going to build more content and then sure like maybe down the line you're gonna get someone greedy that's gonna see it and be like oh I can make a game in six months that means I'm gonna make 10 games a year with like a bunch of teams that's technically that could happen but that's not what we're kind of aiming for we're just aiming for you still have the regular game cycle but you're building more content better games more fulfilling experiences and ideally Europe changing the things in the beginning so you're like alleviating the crunch and you're kind of making the ideas kind of stick a little bit better and that's really the true go for it but like any kind of tool there's you can take it for good or you can take it for bad so the first example I want to talk about is basically the simplest kind of

tool that I could come up with which is making baseboards and crown molding so the usual workflow you would go and your Google baseboards and then you find out the profile that you want and then you're like okay cool I like those and then I want to try to build this for my game and then you bring it in and then you bring in the image and then you go with the Tres tool and you go boom boom boom and you start to kind of going and then nobody got time for that this is just a giant waste of time so the first change that would do is that will look at manufacturer specs so like a lot of these things that we're building is out in the real world so you can just go directly to the manufacturer and get the exact specs that you can you need and then you would do instead of tracing things by hand you would build a trace node that we have and then you just build the line and then you sweep it and I have the same kind of tool that will do the same workflow and because it's procedurally instead of every time you need to make a change or every time you need to make a variation you just have a system that you have you don't have to go and kind of fudge the points

by hand again you just kind of feed it more things and then the system would just kind of go and make things which is a better way of working my opinion so now you have this kind of procedural straight line for this stuff we want corners right so we want the kind of concave and convex versions of it so you can make a modular side out of it and then again the worst case scenario is someone would just put two of them together and then vert weld them hand by hand and then they'll kind of fudge them and then kind of get it aligned again nobody got time for that so the way I like to do it is I actually clip it so it's kind of like a real operation that's like you just mitered the thing so you just cut it a 45 and then 45 and then you show them both together some people even on some of the comments is you want to sweep it along a curve the problem with that is that that's not generic enough so you could do it with just like a instead of sweeping and along a straight line you sweep it of course a 90 degree angle but then if you have any kind of minut variations or if you want to build a general system which I'm going to talk a little bit further

down the line a mitering approach is a little bit better and work on a general purpose so is keep that in mind towards like if you're building something if you can extrapolate that to the next level of how this can be useful for another thing and not just for the one problem I'm trying to solve that's helpful so now I have a full set so now I basically have a baseboard with a concave corner and a con corner and then because it's non-destructive I can just feed any input into it and then I just get the whole set so this is by far like the argument that's like oh it doesn't take me too long to trace a curve right it's just like like a couple of hours I'll go in it's like yeah but you could do it in seconds and then the idea is like the next step is instead of feeding one input we just feed a sheet of inputs and now any tool or any asset that you need to build you just kind of run it through the pipeline you just have like infinite number of assets basically but it still art-directed it's still controlled it still fit what you want it to look like but all other kind of boring little repetitive work is kind of gone so now

you have a tool and then maybe make this tool and maybe make something else and then it's it's like people start to notice and the way usually works is you make a tool and then you just use it yourself because it's so fast that you'll be like you're the guy and then it's like hey do you need baseboards just ask me the baseboards feed me the thing I'll just run it through my system and then I'll spit it up for you and it takes no time but ideally you want to kind of diffuse that out because then you become a bottleneck yourself and you kind of want to displace that you want to make the tools easy enough for other people to use then you're gonna start basically building tools for other people but the problem with building tools for other people is that you have to convince the kind of task keepers and the producers that it is worth spending the time and now it needs to be a little bit more robust and you have to kind of capsulate the problem a little bit you can't just have like the sake tool that you just know where all the inputs are you kind of have to kind of extrapolate it up a level again much like regular tool development the one thing that I will say is don't argue for time to show it there will always be people saying that it can't be done there's always

always going to be people that's saying they can do it faster by hand if you believe strongly on it just do it so I had a great example I used to work on a previous company and we just had like head topologies and we had like 200 heads that we needed to change the topology on and there's a big argument whether or not that was worth doing and with I was like well if we change to another head then we're in line with other Studios and they give us better tools but the argument was like well there's gonna be like it's like 200 heads it's gonna take you like 2 two-men weeks to kind of go and wheat apologize all of them and it's just like it's not worth the time the real problem was like a political thing so it was kind of being guys so one of the guys Johnny who's a Houdini wizard over at EA basically he just went he was like I just wrote a script overnight and I just batch all the heads and the problem is gone and it's like the next time we went to the meeting they were like okay well this is gonna take forever and psych no it's nice done and then the real problem like surface that it's like oh they don't want to change it it wasn't because of the tool or the time it was because of some other reason but kind of being able to just do it by yourself and being like just take out the whole middle layer of

argument and there's no time and it's it's it's it's worth doing and a little bit of the ask for forgiveness and uh permission kind of plies here and the other thing too is that build a little prototype a lot of these tools can be built with very simple nodes and you can kind of do a proof of concept and kind of show it if you can't look at examples of people doing it there's a lot of people doing procedural ISM now so there's a lot of talks from you aid they'd have a great stuff on SSX how they built the whole mountain generation system there's a lot of stuff of uncharted how they're doing procedural destruction and using Houdini for that stuff and then there's ya just go out and then see generally what you're trying to do and there's a lot of resources either on GDC at SIGGRAPH or on our website the other way to approach pitching tools is as you're buying time and this is kind of a weird mind thing because if you say hey I need whatever five thousand dollars to buy this piece of software but that's gonna save you a lot of time people don't see that as like on a budgeting mindset it's like people are free because I'm already

paying for people so I don't want to pay this much money that will give me theoretically like X more people that are just production like man time it's still it's a weird kind of like non logical argument that you have to have but just try to pitch it as time it's like you're not buying software for then you need more people to write the software you're just buying the software that can just automate the whole process or you're building the software it's not necessarily buying but the time that you're using to build it the other thing to use like buy get the binds from the team so the worst thing you can do is come up with a cool idea about how the Turing system should be built and then you go and you try to shove it down the terrain team's throat that's never gonna work like sometimes you can muscle through it and sometimes you get some people that are like oh yeah I can see the light 90% of the time you need to kind of work with them throughout the whole process and kind of get buy-in from them as you're developing the tools so then they feel vested in it otherwise they're just like even if the tool is good and it's better and it's faster just a logical argument

of like I didn't build it I don't know who did this it's not good enough because I didn't come up with it and those are again in the kind of argument land it's it's subtle differences but you're gonna run into it and I know that card lets me personally very logical very just like hey just clearly this is more efficient we should just use this and I'd be like oh you get kind of more emotional arguments with it the other thing to say is like just show how fast people can do it shows some tools to some clients and they're like like this just like shows me that we're working way too slow and then just looking at some of these tools as like see people just like drawing curves and then whole cities apply and it's like Jesus Christ like we're placing parking meters by hand and like we've did that in games it's like yeah but but I need control where the parking meters are because of like the way it's frame it's like no you don't like just please just just just see how fast you can do it and then people are gonna be like well like placing a parking meter is just like really fast I can do it like super fast I can just like clicking and then I'll drag I'll drag drag drag and it's like yeah but if you want change and if the street move now you have either floating

parking meters you have to go in and every time someone changes that you have to go and update it so there's a little bit of that again like oh it just takes me five minutes but it shouldn't take you any time and then that's it's those little five minutes is what bleeds production dry it's not like the big tools that take months like the big tools that take months usually are the things that kind of solve production the little paper cuts is what kind of destroys you the second example that I'll go into is this pipe tool and this is basically like something that you want to hand off to an artist and you want to kind of extrapolate things out a little bit and then you just now you have a tool that you just take a curve and then you get a pipe out of it and this is kind of just a traditional problem they have the thing I have is you want to break down the problem a little bit so you want to take the curve you don't understand how that workflow is and then how do kind of you would come up with the solution for it so the way I would come up with this is you take the curve input and then you separate the corners from the straights and then you will build the straights and then you build the corners and then you assemble it together and you have a pipe so fairly kind of straightforward

workflow so let's go that hand by a hair of step-by-step so you start with the curve is an input as a nice bonus another thing that people like to do is on line curves to surfaces which is a giant waste of time so basically here's what I like to do is you take the curves and then you take the first and last points and then you project it into the surface that you're gonna do it get the surface normals out and then you stitch that back into the curve so now the kind of connections are always going to be perfect because that's the giant most waste of time thing that I've ever seen but that's one of those that's like we don't know what's like we don't come up with this and it's like we don't know what's out there and we don't know how to fix it so we don't know the problems we have then we want to separate the straight from the corners this is very straightforward basically you just take each line segment and then you kind of like carve them a little bit so we kind of spread them out a little bit and now you have basically the two kind of problems that's isolated and then you model a pipe so the way I model the pipe in that example is you take a cylinder and then you sweep a you sweep a circle through a through a line and then you

put two cylinders at the end of it then I made some little BS screws that just look like something and then you merge the the long kind of pipe part of it with the kind of end caps with it and then you do a little bit of a beveling at the end of it not fancy but really really basic modeling but it kind of from a far away it kind of looks and that's the other thing too that's like if you talk to an artist I'll be like oh no I will like bring this into ZBrush and like manually like do divert the the welding and all this other things it's like yeah but then you this little tiny pipe on the corner of the ceiling that no one's gonna see does it need that much attention sometimes yeah and then you that's the part that you can have to go through is like do you build something procedurally or then you give that freedom back to the artist and be like okay fine just build the pipes give me the pipe pieces and then I'll manually place those so those are the two approaches so then I'll just place those just take each of those little side like segments that I did and then I would run it through the little pipe generate through that just kind of strew the the main salon and it puts the two things at the end and inputs it does the

bellami and then the corner is actually the same tool because you just have something that sweeps a circle along a curve and then you put at two caps at the end of it and then you you add a little screws to it so now you're done and you have a tool that is kind of fairly straightforward fairly logical anyone should be able to go in and do it but now you have this kind of cool system that is like this pipes are solved and now for your production of the rest of the game it's like you can you can be the pipe guy or a dealer you hand this off to the artist where you just be like okay here's your pipe tool just draw as many curves as your heart desires and then you just run this through this tool and then you have your all the pipes you need so in the last stage which is the heart stage is now you kind of have maybe we did this pipe tool and maybe like five environment artists use it and that's all you need because you don't have that many pipes on your game but now you're trying to build these kind of like larger systems like a building generator or like a room generator how do you get into that so the way I like to think of it is basically you have this thing connected subsystems and I used to have a friend any problem that you ask him you'd be like you know man the spaceship is a

complex problem there's a lot of moving parts it's not like you can just it's it's it's it's complicated but kind of try to feed what the critical path is and know what your starting over here you need to get over there build little milestones of failure points that you can abort but then kind of know what's the minimum valuable product that you need to build like any other tool but it applies very much so to procedural ISM the idea here too is that if you kind of know what the whole pipeline you can just selectively replace pieces of the pipeline with procedural ISM and then leave the kind of things that are a little bit more close to people's heart or things that people want to do and then you so keep the eye on the prize so you might be like yeah I built you a pipe tool but that pipe tool was gonna tie with the cable tool which is gonna tie to the room tool which is going to tie to the city tool so you kind of start with little pieces but kind of keep an eye on the prize and kind of know where you're gonna go the other approach for this is that there is no plan B and we either do it or we fail and that's a bold move but it works sometimes so caleb howard from EA was like when we did SSX we needed to do

300 tracks we either did 300 tracks or we failed we couldn't do this by hand the problem was so big that if procedural ISM didn't work we would just not ship the game and that's that's the option and then that that puts a little bit of what I like to say the panic threshold which producers will start because - like procedural systems aren't like regular art that's just like every week I make 5 widgets and then I can see 5 widgets being built every week and then if a week 5 widgets weren't built I know that I dipped little bit procedural ISM is like okay I don't have anything and then I have a thousand widgets and then then that that panic kind of goes away but then it didn't work and then like the widgets needs to change somehow and then the panic starts to creep back up and then they start and then and then finally like the widgets work and then the panic kind of goes away and then the thing system is stable and people kind of but know that like and it's it's uncomfortable to kind of be in those meetings because people are like hey what's happening with this and he's like it's coming it's like it's really close and it's like how close it's like I don't know it's like it yeah it's like

I'll keep building it and then I'll keep hitting another wall and then I'll go through that wall and then I'll go to the next wall and at one point there's no more walls and it works but until that point happens I can't tell you and then people get really nervous and they're like well let's start with plan B and let's start getting some people actually building the widget by hand but kind of try to stick with it and kind of try to kind of show them that it's normal to kind of panic and it's normal to kind of not trust the new system but know that other people are doing it and that it's a proven thing at this point it's not like this weird oddball new workflow the other misconception is that once it starts working people are like procedural ISM everything we don't need artists like let's make a game with like five guys and let's just like it's like software development so there's gonna be bugs on the design there's gonna be things like ambiguity that an artist said that they needed something and that they don't have it or then you're gonna build this beautiful building tool and then someone gives you just bad modular sets that completely break everything and then it doesn't work so there's no such thing as an off-the-shelf solution yet and that's the part that we kind of

need to build as a community and kind of start getting kind of more standards and kind of building it to that we get to the point that's like everyone wants to build their own procedural system because there isn't the one building generator that's of like just like just works for everybody so everybody has their own things like I want to do a premium building so my buildings are completely different I want to do an old-west building so it's completely different until someone comes up with the PBR of buildings that is like you just can't mess this up you just give it a curve and then you saw like which style you want and then you get better curves than anything and that's I think where substance did really well towards just like you just have a library of really good materials yeah you want to make gold go make gold but the gold that they give you is pretty good so then people start to kind of go down that kind of pedestal of like my gold is better than it's like yeah do you really want to spend 30 hours building a procedural wood shirt go and spend and do that but like 90% of the time you can just find something that is already done at a quality that you like and then kind of move on so for my final example I have this modular set which is as you can see it's very similar to a something I

showed earlier today but it's kind of like the extrapolation so the problem that I have in this one is I have something that is excised and I need to fit it in something that is why size and then the way I usually start with is like okay I'll just place it there and then I'll place another one and then the way we do it by hand is like okay I have a wall segment I'll put wall segment wall segment and then I'll put some weird oddball little corner shape that didn't fit because people didn't really apply to the grid or it didn't just kind of fell off and now I have this one custom piece that it's like a weird corner and ideally what I would like to do is that with like the procedural system to calculate how many things can I fit there and then distribute the weight of that scaling slightly by all the pieces so then you get a little bit like a 5% squish on each asset but it's not noticeable and you don't have all these weird little custom end caps that you have to figure out how to do it so again the math of this this is the only math of the whole talk it's fairly straightforward so if you have something that is too wide and you have something that is one why'd you put you figure out what the size needs to be so you need to

multiply it by two so that's just like a basic kind of ratio what you have to multiply it and then if you do it one by one it starts to get easy but then the real problem is when you have something that is like 360 7.50 8/8 and it's like okay then I just want a system that deals with that and I don't have to want to worry about that so if you have something that's 5 you divide 5 by 2 and you kind of figure out how to do it and then the other problem they have to do is the number of segments so how many of these because I don't want to just get one really long segment I want to figure out how many of these segments can I fit in this wall size and then I want to kind of subdivide it so you're just divided by 2 and you kind of round it so you got an integer and then you figure out the individual size of those so you kind of have the individual piece it's gonna be a scale of like a smallest small scale of what the original one was so you kind of get the full watching my job like how do you call it just the whole don't whole math and it I know like you guys will go over and kind of look it over but it's it's really simple math you just kind of you have the wall piece you figure out how

many widgets you want to put in it and then you figure out what is the Delta of the full-size versus the widget size that I got divided and then it does just how much you need to scale things by but once you have that and now you have a system that you can do this and then it was just like oh crap that's magic but it's like it's like three little equations but it's really like now you have a tool that just places things along a wall panel and you don't have to deal with it and then things will stretch up to a threshold and then they'll kind of move to the next as a stretch but on the opposite way and then Oh kind of start growing and then if you put that with the real asset you get something that looks like that and as you can see it's like yeah you got a little bit of a stretching and you can see it really bad when it's moving but when you kind of freeze it it's kind of imperceptible so now you have this thing that you can squish but I also have a way to kind of miter things right so then I have ways to kind of make things squish this way but then I can also figure out how to put corner pieces so what this means is that I can have a room generator so I can just make a

thing that just takes a box as an input and then I can make rooms with it with really basic math and it just works and you can see how this starts to extrapolate so we started with the baseboard that I was like what the hell is this baseboard for but now we get to the point that you could see how it's wrapped like this to a building so right now this is the first war instead of kind of having things facing inside I'll face them outside same kind of technique and then you just build things up and then you do the same thing when you're building things up to where it's like you divide things equally and you kind of put the first floor in the last floor and kind of squish the middle parts but getting to this stage to where it's like now you have tools and that's how people work that's a hard kind of mind thing people need to drink the kool-aid and they have to be like a min I get it that's how I want to work I don't want to place things by hand and ever again so there's the costume both and kind of getting to that there's a lot of ego of people being like I'm really fast and unreal and there's a bunch of tools and I can just blue blue blue and it's like yeah I'm sure you could but like getting to that stage is kind of really important you're gonna get pushed back and the best way is kind of like use

these stalks use other thoughts that people are doing with procedural ISM and kind of help drive that home and kind of show the light a little bit as if were so the other thing too is that I like to think about this vertical monopoly - where if you have one drive with one driverless car in the road that's great because you have something that's automated and is driving by yourself but the moment that you have a person driving the car now the automated system needs to figure out how to drive around the bad driver and there's all these other kind of problems associated with it so the real system is like when all of the driverless groaned the whole pipeline is automated what does that open up so it's like when you have the whole thing that you can make a room you can make a building you can make a city you can place anything what does that open up like now you can have massive worlds or you can have small worlds and lots of them so getting to that stage that like the whole pipeline is automated is kind of like the golden Grail a few studios are getting really close to it but you're gonna start to seeing that kind of paradigm shift of just like now games are massive and there's just a lot of content and they're all triple-a and then a lot of people won't know that they're being used procedurally which is the nice part so it doesn't have that kind of stank on

it that is like oh it's just fractals and random seeds so that's that's there's a little bit of so just over all the the pros and cons of kind of going procedural so it's non-destructive so you can kind of change your mind because that never happens in production you have a whole pipeline that is as much automated as possible and the really nice thing is that you can have a process that is parallel so now you don't have this waterfall pipeline anymore you can have the character starting and you can already start reading and animating and know that once the character is done the system would just pick it up and transfer weights and everything is done so now you don't have this like a lighting get screwed always because they're just at the end of the pipeline and then they're just going to eat the the little crunch at the end of it it's like no they can start lighting in the beginning because everything is already in place and we're just kind of building the the shell for it but as things kind of resin everything kind of comes together the cons for it is that there is a longer rainbow there is an investment upfront that you kind of have to go through it there's this perceived loss of control which is perceived in

the beginning and there's a lot of people that are like yeah I don't have the control but then really it's like well did I really need the control of like manually placing the screws on the wall no like if I just draw a spline on it that's enough control for me and then it's new and different and that means it's scary and that's something that you kind of have to be cognizant of as you're kind of pitching things through that just on that emotional level people are like I've been doing games for 20 years and that's how I've done it and we've always done it this way you're coming and he's telling me that there's completely different way of working I don't like that I want it like there's job security and me knowing the pipeline and me knowing how it's done if you come in and displace that that's scary so it's something that you kind of have to be aware of and it's work with them and it's as much as an emotional coaching as it is just like a production building so then as you kind of like if you hitch that level of okay now we have a procedural system and like we can build anything that we want it's just iterating and basically like each of those modules can be better you can build a better poly reducer you can build a better quad remeasure you can build a better transfer of weight so you can better build it better you being so

you can kind of start driving that and then you can also start tying you a I driven systems to it which I have a talk in 10 minutes over there which is basically you now you have this black box that takes input but now we have this other thing that can make inputs to drive this black box and then now you start to get a really cool interesting approach that you have the artists are kind of curating the content more than manually generating them which is a little bit of different but it's kind of Democrat Democrat ties is it a little bit so it's a little bit of if I can just draw a paper map and make a level out of it now QA can build maps like that you don't need to be a master world builder and unreal wizard to be able to make a map anyone can make a map even if it's just a white box and that gets us really closer to where we need to be which is 3ds to expensive to make right now there's all sorts of fields that are trying to get into 3d with like photogrammetry and like historical preservation and archaeology and all these other things that are trying to get it but is just too damn expensive because it just takes too long so

getting it to where building content is cheaper you're not gonna be running out of jobs there's plenty of 3d work to go around here's just a matter of kind of distributing it from being in the games industry to going out wide which i think is the next evolution for for our kind of crew so that's the rest of it come say hi tomorrow we have a Houdini training day so I'm going to be talking from noon to one and it's gonna be over at the Marriott and then we're gonna have booth talks I have two talks every day and then you can just come by I have another talk right after this so I don't think I can hang around too much for questions but I will but then if you can't grab me grab me over there and that's all I got and how long do I have ten minutes twenty okay so we have so we have two options choose-your-own-adventure we can do QA and we can just chat and kind of talk over whatever questions you guys have or I have more content that I can just

ramble for another 20 minutes and show of hands who wants questions who has questions okay then I will go on to bounce around and then just lightning around more okay bones round so these are the things that didn't make it to the talk and then now I just got to talk as the talk is over so another example that I wanted to talk about was dirt skirts this didn't really fit my perfect paradigm of you have a perfect little thing and then you move on and you kind of like it's a sequential sequence of examples but it's just kind of like another thing that people do we lost months of work doing this in a previous title and then the idea here is you have a piece of geometry that's colliding with another piece of geometry and you want to make a little trim because that line is really ugly so the two inputs to the system are gonna be have a piece of terrain or a piece of flat geometry and then you gonna have a rock that is a piece of just another piece of geometry so the way we do it is we boolean them together and then that

basically with a new system you get a curve that is the boolean that is basically where the two things meet so now you have a piece of geometry that is kind of like unified the rock with the terrain and then you can have a curve that is where the two of them meet then you just sweep some geometry along that scene we convert that to a volume so now we kind of have like a little donut that is like the kind of skirt for it and then we do an intersect with that Union that we had originally so now we kind of have a just the skirt part of the the kind of two parts of it then we offset that by the normal and that just kind of makes z-fighting doesn't hate us and then we add some vert off at the edges of it to kind of fade it a little bit and then we add UVs procedurally by just doing some like planar mapping because it doesn't really matter because you're gonna put some noisy texture here anyways and then you apply the material so that's kind of the the secret of building dirt skirts and now you just have a system that it's like oh now you just have you just have geometry and then things and then just

make dirt scores for the whole map so we started to do that previously we didn't have this boolean tool so it kind of didn't work as well but now with 16 we have a lot of stool new stuff that makes this super easy and then the other thing which is my main talks and other things is basically this crazy stuff that I'm doing called vertex animation textures some who has seen any other stuff online all right sweet so this isn't sixteen Houdini 16 and it's basically a way of gaining geometry caching and Ben is gonna talk a little bit about that too his talk so I'm really happy to kind of like as talks are happening at GDC you already have a DCC app that can do this stuff so it's a way of getting geometry caches into a game engine using textures so the way that works is and we had another skip button casing right now the time but there's four different ways of doing this caching so there's the we call soft which is the same topology all the way through the pipeline so it's like cloth and soft body dynamics and that kind of stuff rigid which is moving

chunks so it's basically you have a bridge your body destruction fluids which is like a changing topology and then sprite for a camera facing cards so I'm gonna go a little bit into those so the the soft basically you can get this stuff so you can get cloth simulations you can get oceans you can get soft bodies and you can actually get skin meshes which is really cool so the Uncharted guys give a little peek on this and the PAR gonna have another talk this week on how they did the chickens in uncharted 4 by basically using this kind of technique and tying it with their particle system because in the particle system you can just spit out static meshes but then if you can have a shader that has animation on it and I have static meshes with animation just running around the world and you can just spit out hundreds of these and they're fairly fast so what the hell is happening underneath the hood so the way we stored the data is we have the the textures which is actually a really good format because you have every pixel has red green and blue component which means we can put a XY and Z position in it and then we just basically store that

and it's uncompressed and then we make the UVs basically be lined up horizontally on the mesh so you have a second UV set that just kind of like it stripped out horizontally and then on the texture side you basically just put every pixel is a point and then every frame is a row so then the shader actually needs to be super simple which is just you take the second UV set and you just scroll down and then it kind of does the piano roll thing where it just kind of keeps setting the world positions and the normals every frame which works really well so this is the unreal shader which is not super complicated so the kind of interesting part is in the bottom to where and this is online and I have it on my Twitter but basically it's the hard part is just calculating the UVs and kind of scrolling it down but that's really it and then you just sample the two textures on the vertex shader and then you just set the position in the world position offset and then you set the normals and you have to do some shenanigans in unreal and specific because you have to pass it through the custom UV Channel so it gets calculated in the vertex shader and then you pick it up from the vertex shader and pass it

to the pixel shader so what else can we do with this so we can do destruction and instead of storing positions and normals we can store positions and rotation and your two textures so now you have this kind of same they have two exactly the same simulations just export a different way so one way is using the texture animation the other way is using joints so the same kind of system that the way the artists work is the same and then they just output the data the way they need to upload it but the cool thing is if you're going to try to do something with like 3000 joints that would most of your consoles but with this thing we can kind of get it and it's super fast and it just it just works and the textures not that huge because you just have three thousand by a hundred texture and then you can kind of fudge it a little bit or you can basically get I got 1k by 1k texture so for this we basically store the position in the quaternion and on the texture and then we store the pivot of each chunk in the vertex color and then all of the chunks verts have the same UV so

basically because they have the same movies that you sample the same position and rotate so they kind of move as a single unit and then they basically have to do just the the the movement of the position and then the rotation but because you have the pivot you know how to rotate it from there and it just works and then let's see so then the next one is the sexy one which I really like because there's really no way of doing this besides going into a lambic but then Alembic has its own problems of basically recalculating the geometry every frame Ben is going to talk Friday a better way of doing this so like I always like to say this is the artists solution to the problem like ideally you get a proper graphics engineer that can Hotwire the data into your graphics card but in our cases that were just kind of like bootstrapping things together it's like yeah we just write it detectors and read it because it works on people and then the graphics programs like no textures are horrible why are you doing that but it's like hey man it works and you kind of move on with it so then you can get things like this kind of stuff which I

really like and it's just you can change the topology and it just it opens up a lot a lot of really cool effects and what we're doing here is is the same thing as we did before it we're restoring the position and the normals of every point and you have the option of an additional color texture if you want to get the the type of that color from the simulation but then you store a cloud of triangles which is basically the maximum amount of triangles you could get in your whole sequence and then every frame you just set the position and on or most of the the quads and then if there's no position and normal so get collapse down to zero so every frame the kind of the the mesh gets reassembled and because it's already there and you're not really creating geometry every frame it's actually really fast and it's fairly cheap so like for this particle effect I have something that is five thousand tries and 99 frames so you can see I have I need to have uncompressed textures because if I try to do block compression on this with just the whole data will get munched but we can kind of keep it uncompressed and it's still like six Meg's for the whole particle I have some stuff to basically collapse the

normal into the alpha of the diffuse so that are the position so then you only need one texture at 6 Meg's and then the mesh is tiny is like 600 K so compared to if you're trying to try to do this in Alembic and have like a 50 Mac cache or something more massive yeah this is fairly efficient I haven't run numbers on it yet because it's too kind of bleeding to new I just finished getting the bugs worked out Saturday that's why it's not on the main talk and then the last one is basically sprites and then this is if you want to kind of get GPU particles but you can actually sim it and get it art-directed and kind of get it doing exactly what I want it to do and then you kind of bring it into the game and kind of have the same look so what we do here is similar to the other one too where instead of storing this triangle cloud we just store quiet cloud and then we basically move the positions of each of those to where they need to be and then we just make the shader be camera facing so it's very all the techniques kind of build on top of each other and then yeah so this is the kind of whole

shebang of the things like the sprite is not on this one but you can see how and this is me making bad art because I'm like a lazy artist I went to art school but I became a tech artists because I'm lazy and I don't like spending time making art but if you get like a real proper and you can swing by the booth and we have better examples of like Magnus doing some amazing like lava do you guys might have seen online this if you give this to a proper artist and knows what they're doing with you're gonna get some amazing stuff and that's really the end of it alright so now I'll take questions and then we can go over the next thing thank you again second questions hey so great talk just I feel much less alone right now okay this is absolutely everything I think since many years so

it's really cool to see all these people listening to this and the the sometimes when y'all check out it's in your company you feel bit alone and yeah it's all the the thing that nobody believe its magic and okay it's not a ball so it's very interesting and I feel something is missing a bit to this talk maybe anything because it's just a short but it's most of the time we try to solve problems we try to automate things so there is some kind of defined problem and we try to find a solution but also sometimes we find something new yeah we answer a question nobody asked and that's very interesting too because it's not always trying to fix a productivity problem sometimes it's also art and it's another way to do art it's just using anything CODIS Creole its software's

anything to do art and what do you think about this I think it's awesome and I think that's where it gets into really interesting land because you move from tech artists being this kind of firefighting unit to being the people making the content and being the people that there are the only people that could pull this off so when I was in production I was that like I was the river guy and I was the kind of heart problem guy so and it's still in the problem space but like if you can get to the point that that card is driving that and they'll be like hey there's this crazy new game that is possible only because of this new system that I came up with that is that that's absolutely like I come from a broken down production jaded environment so I'm just looking at the next problem but yeah seeing it as a there's just a wide open field and now the tech heart is it's kind of driving it and it went into another thing okay yeah absolutely no I that's where I want to get to to where that card is are the ones making the content and then we're the ones kind of pushing you forward very very good point Thanks yeah everybody thank you so much

and then come talk to me afterwards they're in the booth and then it's up next