Low Poly Modeling: Style Through Economy

[Music] all right so let's get started uh my name is Ethan red I'm here to talk to you guys about low poly and achieving a sense of style through economy uh so let's start with an introduction I'm sure you're all wondering who I am uh I'm working on a game called blazing Legion ignition it's uh in a retro sort of Sega satury looking 3D style and uh it's currently in development mainly just me and two audio guys and I've I've released a game called neo1 which also is in that same vein uh a full 56,000 Square met explorable uh Island uh yeah that's up for download on my itch page and this other game rad Road Rally which is about the that vom vroom sound you make when you're six playing with Hot Wheels uh so yeah just like fun owed to Childhood basically it's one of my first games kind of janky but a lot of fun uh so yeah just getting

right into it oh wait actually one more um I worked on one of the scenes for the Dome video at Panorama the concert this concert series in New York uh I did that last year a lot of fun I worked on two scenes one of them actually made it into the video which was super cool uh you should look that up on YouTube if you can but now actually getting into our topic of what is low poly uh low poly is what I like to call a quantifiably qualitative in that there's there's no hard definition for it but I like to think of it as the discipline and aesthetic of surface economy so basically describing 3D or sometimes 2D uh surfaces in as little geometry as necessary to get the idea across um and low poly is essentially a low Fidelity art right so we're thinking about efficiency uh more than we are Fidelity we're not trying to accurately recreate objects we're trying to get the

idea of them across and again it's subjective uh there's no hard poly limit where you are or aren't low poly but generally for me I like to think around AK per asset is where you cross the line if you can see angles along curvature that's that's probably low poly and we've seen sort of in 3D an evolution right of moving more towards High Fidelity uh Graphics especially in 3D because we can basically uh especially at the higher scales of AAA develop we think of higher Fidelity as better up until recently where we are seeing sort of like a Renaissance in low poly art uh for one big reason I think money right um independent developers we don't have the same resources that triaa devs have to play with we have different schedules there's less of us working on a project uh we have tighter time constraints like things are different and it's a bit hard to go for a high fidelity

looks when you are worried about literally everything else so lowly has um it has a place in modern development besides just a last second optimization step for bigger games so modern benefits first one being productivity uh low poly has what I like to call a lower existential barrier uh than other uh ways of handling 3D and that you can get your idea out and across much faster than than you can like a highres sculpt excuse my computer um faster than you can a highres sculpt or uh just generally a richer asset um and you can get up and running faster which is great especially if you're into prototyping or trying a lot of ideas and being quicker produced allows us to increase our scope as smaller developers um if you're trying to make characters that are completely fully realized photo realistic like actor

duplications you're going to spend a lot of time on that where if you're an indie developer maybe your time is better spent making more assets and spreading that detail across and uh making more instead of making one thing and like really digging into it and a big way that we accomplish that is in our workflows right uh what I tend to use is the one at the top which I call low poly first where the low poly aesthetic is sort of baked into how the game game looks so the first run of the asset sometimes will be what I use in the final production game but if I miss it's quick to iterate on and I can try out a bunch of new ideas and keep moving and then even if I want to go onto a high poly asset at the end um what you usually see with low poly is we'll start with a uh high res asset and then uh there'll be a manual reduction of it or someone will run a decimate operation over it which is bad

uh best case if you run a manual reduction and you go and make a separate mesh um you can clean up uh a nice smaller mesh for it but you did the work twice and if you're going to end up using the low poly mesh sometimes it's better just to go with that first uh the worst case scenario is the bottom which I call afterthought optimization where like you want the highres uh asset and you realize that you can't really use that so then you go and run a decimate on it or whatever just to clean it up and then you realize that's trash and then you have someone come back and make a low poly asset from scratch and you've done three steps when you could have just done one to get to that asset and this is even worse because if you don't like the result of all that when you see the final silhouette of the low poly acid you'll have to go back to the first step and make a new high poly one to change that and then you go through this whole chain again and it's completely inefficient so low poly first thinking is a great way to save time

and then on the other end we have computational efficiency which isn't existentially necessary these days because we do have Hardware that you can just throw a million verts at and it will run and run smoothly but that's only when you're working for let's say desktops or newer consoles if you're producing for small screens efficiency is still very much a consideration and you want to keep that in your mind when you're designing things uh also for audience purposes if you want to reach the next billion users like Emerging Markets or reaching people who don't have access to new technology or are playing on older computers that popup is annoying me um if you want to reach people playing on lesser hardware and just generally uh broaden your Market low poly is a great way to lower the threshold uh for who can access your game and then lastly having less overhead um if you're rendering things

twice for example for VR or if you are doing AR and then you have other processing going on graphically having less geometry to worry about uh saves the GPU to worry about other things and you can spend that computational time on just other stuff so now that we know why we want to use low poly we want to start thinking in low poly basically and there's four principles to this I think one is stylistic compatibility if you want to make a game that looks like real life this is probably not a good match for you because fundamentally low poly is about reduction it's about taking out the details that aren't necessary to get your idea across and leaving and accentuating the ones that are so if you're working with something that's angular or generally sharp or even uh something like uh wood block paintings or I don't know cubism or uh origami is

a very classic example these are things that translate to low po very easily and naturally and are sort of like a leg up in your design process you can break away from that but you're cutting Against the Grain so this is something to consider at the onset of your project the next pillar is resolution Consciousness and this is where I think a lot of people slip up with low poly is um we sort of just will think oh just reduce it as far as you can or reduce it till we get to our uh Benchmark and then you're done but when you do this there's there's no human touch to it and your viewer will pick up on that and the general look of the game will suffer by having a sense of consistency through how you handle the polygons uh that make up your mesh it's it allows you to push the style of your game and we'll get into this in detail the third thing is considering silhouette at all times now if you're already a 3D modeler this is something that's already baked into your brain because you know you need things to read like they are from all angles Les but

with low po this is even more important because again we're stripping away detail so we need to make sure that the overall is very strong and ultimately we're approximating things and we're going to dig into this in just a minute so we don't really need to think about conveying things as they are we want to convey them as what's important uh about them and then last pillar is uh either supplemental or old school rendering tricks that just sort of help you sell the style you can fake things through particles shaders textures or even incorporating other Loi or NPR techniques and then that allows you to achieve those looks we talked about earlier like looking like uh origami or wood block painting or whatever so starting with resolution uh the easiest example for explaining resolution actually came from calculus believe it or not when I was looking for examples uh which surprised me uh so if we're for for example our

data that we wanted to portray was the red area under the blue curves and we wanted to show that red data in a series of finite bars uh the if we had an infinite amount of them we could actually show the data as it is which is the uh leftmost graphic but then was as we reduce the number of bars generally we see that we're approximating rougher and rougher with 32 of them we can sort of get the idea if we added them all up we have sort of closest like a close representation to the data and this is High Fidelity this is when something is very close and good enough to represent something if we cut that in half we notice that there's a lot more white between the blue and the red and that's uh sort of showing a margin of error but we still understand that that's a curve and we still can see generally what the data is so here we have a low Fidelity representation of the the area under the

curve but it's still good enough we get the idea and then if we move further to the right than that we notice we just don't have enough data and we have garbage this is what we want to avoid so it is possible to reduce too far and it is possible to present too much with low poly we want to find a happy middle ground between the 32 and the 60 uh 16 example so extending this concept to shape um if we instead of think think about uh the cuts instead of as a verts around a circle like if we're trying to approximate a circle here 16 vertices is indeed enough to show that it's a circle like we don't even really have like a name or a common name for a shape that's 16-sided uh when we go over to the right we have an octagon which you know it's kind of round but not really you go further it's a hexagon and now I'm really pushing it and then when you go further than that it's a square so at that point we've reached insufficient data to represent our Circ Circle but

from the hexagon upwards we kind of have enough data to show our viewer that we're portraying a round thing if we go to three dimensions now and we think about Rings cutting along our uh sphere that we would like to approximate 64 Rings pretty much look smooth I didn't smooth any of these models uh 64 you can't really even notice that's polygonal uh 32 is indistinguishable from 6 four so if we wanted to just for optimization sake we could cut out half of the verts well half of the Rings that' be more than half the verts uh of this model and the end viewer isn't going to notice we can cut in half again and end up with 16 rings and it's still sufficiently smooth to show that we have a sphere and then once we get down to eight we're starting to see the banding and then once we get down to four it's no longer round but I think this illustrates the concept that you hit a point with a curvature that's good enough and you've made your point

once we get down to three it no longer has any semblance of being a sphere without context which then brings us to the topic of what I call atomic size um if you notice all of these uh spheres have the same amount of rings as examples on the last screen but these all have a rigid atomic size um that meaning the smallest polygon on each of these models is the same size which then leads leads the scene to have a sense of scale and it allows us if you notice all the way to the right I'm using that janky four uh three and four ring uh sphere and it looks more round in this context simply because we're communicating that there's a finite amount of detail that we're spending in this scene and once we get down to smaller scales there's less to work with and this has a lot of uses and is one of the best ways to give your low poly art a cut a above a lot of what I

see is by having this sense of order and cohesion and when you're rigid like this it kind of sells it when you're working with looser approximations of objects but you don't have to be completely rigid with it in this example you see that the larger the largest sphere actually has the same uh same amount of cuts as the next sphere to it uh because the atomic size is more of a lower bound and that takes me to my next point of each U atom each model can have its own local atomic size and you can play around with this as a compositional tool and I illustrate this here with two scenes that both have 800 triangles in them and the triangles are basically just spent differently so if if you think of detail as more of a commodity than just something to shoot for we can start playing around with it artwise and uh the first thing you'll notice as like a difference between

these two is like fitting or how lo How Loosely or tightly we're approximating this data uh the dispersal scene on the left is very rough fit of this scene and the accumulative scene on the right is more accurate in that we have more detail in the individual elements of the scene um the way we do that while keeping our uh our poly count the same is by playing with the atom size uh the atom size in the dispersal scene is global where the accumulation scene has a local atom size Tech usage so The Rock has um different size polygons throughout it than the grass then the tree then the leaves where on the left- hand side uh generally we try to keep the sides of the polygons comparable this ends up allowing us to do different things in the two scenes because on the left- hand side it's very dense and rich and Lush and this is

great for if you want to have like a forest scene or you want to do a city and you don't necessarily need detail you just need lots of things to look at um the sparse density of the scene on the right uh allows you what I call Innate composition where the scene on the left is more muddy which isn't necessarily a bad thing that lets your eye have something to look at at all times no matter where you look at the scene there's some grass or there's a tree or there's a rock whatever on on the right hand side we have Clarity which is great because it naturally draws the eye places and you can use this in level design to draw the player to uh areas where you want them to go without using text or dialogue or whatever or you can use it just as a way to convey story if you have things where you want to like have the player look at and read or whatever um and it's something you can play with on a design level and it's something I use here uh this is a prototype I worked on for like I don't know two or three weeks or so but it was like a generative uh space

game where there's 200 uh star systems and each star system has 12 up to 12 planets with each up having up to three moons and each planet has its own distinctive flora and that's could easily become like a huge project but by using uh dispersal techniques and having the models composed I was able to make these like Rich environments very quickly which allowed me to play with big ideas on a very small scale and then I do sort of a hybrid approach with my game neo1 where I used accumulation techniques on the hero assets like the big trees and the cliffs and such uh to sort of draw your eye to the interesting things and then I use um what I call breadcrumbing by having more detail sort of Hots spotted in areas where I want the player to go because I didn't want to have any sort of like text or dialogue or whatever guiding the player where I wanted them and then on the other end I use uh dispersal techniques to litter the environment and populate it with stuff because you're in

this Lush alien Forest I want it to feel like there's plant life everywhere so just having like a tree here tree there and a rock that's not going to do it uh I wanted to have uh small polygonal grass everywhere there's uh there's little flowers there's berries there's Little Gems or whatever and it gives you something to look at no matter where you are and also by using dispersal techniques a great benefit of modern uh game engines with this sort of look is you can catch light and Shadow so it makes the the ground of the forest look much more interesting because it catches that nice moving lighting and uh just overall adds a little bit of of detailing to a a sparse world so going on to silhouette uh silhouette Consciousness is key to good low poly because things need to make sense right if we're constantly removing details and things that we would identify objects by we need to leave what's important to

recognize them by so the first thing you want to consider when modeling something and trying to be conscious of the silhouette is how legible is the overall shape both on its own and in context um the second Point you'd want to think of is how should light and Shadow interact with your object a really great example of this is actually the chairs that you guys are sitting on where you could represent these chairs a bunch of different ways uh but if you want to accurately like sell the shape you're going to need to use geometry instead of textures or Alpha to model like the bar that's holding the cushion so that way Shadow will shine through it or maybe you could use planes with a texture of the bar with Alpha that allows uh lights pass through but this is something you want to consider because light uh interacting with the form of the object is crucial making like the image work and then the last thing you want to think about is are you over or under selling curvature it's really easy especially coming from like a sculpting or high-fi background to

think that round things need to be really round but again like I showed earlier there comes a point where you've done enough and maybe a couple uh 30° angles is enough for a turn instead of actually subdividing that and then on the flip side you don't want to get too reductive to where you're just making really blocky things that shouldn't be and an example of this is a hand uh from the cockpit of Blazing Legion and the 800 Tri version is the actual asset and this was made first not the high poly one by using clean topology or relatively clean topology you're able to subdivide and get a highres asset if you want it but if you don't by considering what's important about the model I can have different striations that I can use to reduce if I need to for performance reasons so if I wanted to convey just this is a hand in a glove with a sleeve over it I don't necessarily need fingers for that so when I step down to 400 I

can sort of block off the three fingers and still have the one uh index finger available for like trigger activities or whatever while still keeping it more or less deformable if I don't need a trigger finger maybe for the other hand where it's just like holding a uh holding a joystick or something you can actually get rid of that other finger and then reduce a couple of edge Loops out of that get rid of some of the detail in the sleeve and now you've cut it almost in half and now if you know the hand doesn't need to animate you can just lock it into a fist and cut even further from there so by thinking about what's important about the shape that you're modeling uh you can make smart decisions and stylize it so that it makes sense now a bad example of what you'll see a lot is just people decimating a model when they need to get more efficiency out of it and this is ultimately I think a bad way to do it because as we decimate further we notice that our geometry gets harder and harder to work with and maintain and also gets less and less deformable the uh 256 try

example if you Tred to rig that which is have horrible results uh the 172 I really hope you're not intending on moving that at all but yeah and uh from here you can still uh subdivide up obviously because it's the same mesh but I really want you to pay attention to that this is bad right this isn't really doing low poly in my personal opinion there's got to be a personal touch to because it's holistic we need to decide what is and isn't important we want to think reductively about what we're portraying and make a decision about it not just you know cut down based off of angles and numbers and then this is sort of like a lightning round of just like tricks that I like to use that help breathe life into these techniques and also give like a a bit of authenticity uh to the look so something that really uh complement low poly modeling in my opinion is low resolution rendering because the two

sort of came hand inand with each other back when there was there was less uh computing power to work with all around so these are two prototypes I worked on uh on the left is actually a prototype for blazing Legion that was made in like I think a week or so and uh rendering really really quick bad models at very low resol resolution really helps to hide the errors and general lack of detail if you make any uh on the right hand side it's sort of like a starf foxy schmu type deal and it ends up looking more like a sness 3D game and it forgives the uh the pop in that you see at the end that's happening with the level generation because it seems like a hardware constraint because like your eye knows to interpret those things together so this is one of those techniques that just kind of goes well with low poly if it works for your art style I'd say at least experiment with it another one that's also kind of popular these days is using color palletes as textures um um sort of my signature with a lot of my work is I'll

use the uh Sega Master System palette and because I just really like the colors but also uh this is a really great way to batch materials because a lot of time I'll actually have two or three materials in an entire game because they're all using the same colors and that saves a lot of rendering time especially when paired with a reduced uh polygon count so that's something to consider and it's a nice stylistic touch especially if you're going for an unlit look you can get a very nice like a sort of sticker Vibe out of it like the little dude on the left then uh playing with normals is also something you can do uh it's great for doing a cheap glass or uh fluid or anything that is semi-transparent and you see like the back of um I use it here for like showing a glass jar and a prototype I'm working on and and other prototype um it's the helmet of this little astronaut dude and you can

combine these techniques with this one hold on uh this is a scrolling UVS which is another thing you can uh you can do that's cheaper than shaders sometimes and you can also batch this but anyway uh by combining these techniques you can get effects like this where it's like a like really Sega Genesis Sega Saturn 3D looking like atmosphere around a little planet uh just by scrolling UVS and having flipped normals for uh the atmosphere this whole planet I think is under 500 tries so what I'm rendering a whole system of these it's Ultra cheap so yeah takeways I want you to consider stylistic compatibility make sure the look you're going for is compatible with low poly and lends to the aesthetic of it I want you to approximate instead of duplicating objects think about what's important and reduce to that be consistent with resolution either use it uh uniformly to create the

dispersal look that I showed or play around with adom size uh intentionally and get the accumulation look develop strong Silhouettes so that your shapes actually uh make sense and stand out in context and alone cheat hard when possible and please don't hit decimate so yeah thanks for coming uh that's the end of my talk and yeah thank [Music] you uh so yeah I I think I have extra time if anyone wants to ask questions two minutes hey yeah questions yeah right okay um so is there like specifically for organic or for mechanical

things um yeah okay um so to repeat the question who's asking about what to consider when rigging low poly models um it's generally the same as high poly models you want to make sure there's enough geometry to articulate right so if you have a joint like for example a finger you want to have a little bit above the joint a little bit below the joint and something in the middle so that way the Bones have something to catch on to uh so that's if you have like continuous geometry um something I use a lot is actually just intersecting geometry where things aren't even connected and then you can just assign that directly to a bone that's a great cheat if you're doing something like robots or anything mechanical because nobody really cares if they're connected or not um any of the cues yeah y uh so uh uh human eyes are really good at pattern uh Recon nation and uh uh it can really easy to pick up the same model if it repeat too much uh so I think that uh uh

low poly can really synergize with random generation so do youing how we can work like maybe not for the models which you need to rake for animation but for like enir environmental assets like treeses and stuff you can uh so how can we approach this uh if we want to say randomiz generation and train model MH okay so um just to clarify do you want to know about uh how can we provide like a little bit of variation in like smaller assets with low poly yeah and I think it it can also save time for you and save time yeah okay yeah so that's where I'd say it's a good re uh use case for dispersal techniques where you make really small uh course assets and you can just make a ton of them because they take you 30 minutes each and then you just uh you vary those in your generator so just generally like approximating Loosely working quickly and um you know

just make a ton of variations okay thank you any other questions yeah hi and um thank you for the talk it looks like this year at jdc the um subdivision surfaces are finally coming to game engines and it at least my experience it looks like low poly modeling is a really good starting point like even old box modeling is like the perfect starting point to get U elegant control cages for subdivision and I'm asking uh I'd like to ask if you've played around with any of that and have anything um uh subdivision control cages yeah really like when you make a low poly character with just a few hundred polygons uh but but you use Edge creasing the resulting subdivided model is a really really elegant um shape for oh yeah I see what you're

saying like subdivision modeling yeah yeah yeah yeah definitely low poly is great like U like I was saying in the beginning it's a great way to leverage up you know so like working on a low poly c as you're calling it um doing that first you can reap the the workflow benefits of being able to iterate on Silhouette very quickly and then once you subdivide after that you can hang on to the low poly model in case you want to make changes that are like drastic to the silhouette like for example you want to like change like shoulder width or like hip height or something and then subdivide and then you can go over that again with like a sculpting step thank you yeah thank you the other yeah uh hi so um uh this might seem kind of anachronistic since um you know low poly is I mean older consoles like the PS1 didn't have like normal mapping support so I'm

curious what to think what what you think might be uh the potential role for more advanced rendering techniques that newer engines like unreal and unity offer in regards to PBR and uh and uh more advanced kinds of materials in stylized low poy right um okay so the question was where do I think uh low poly could go with um I'm actually out of time uh if you guys want to I'm so sorry oh yeah no worries if you want to talk more I'll be at the uh little room that they provide I forget what it's called and yeah thanks for coming [Applause]