Procedural World Generation of Far Cry 5

[Music] so welcome to for cry 5 procedure all world generation my name is it's in Cali I've been a technical artists at Ubisoft Montreal for about three years now and I joined at the beginning of the far cry 5 project and well I haven't worked on any triple-a titles before but I I joined however when I joined USF I had a secret weapon and my sleeve and that is my experience with the procedural tool development I've been passionate about that for about six years now and before joining us F was working on indie games and that's where I started introducing procedural work flow into my pipeline so today I am happy to share with you the the procedural pipeline we developed for for cry 5 well quick reminder tour

enough your cellphone put it on mute and avoid taking too much pictures if you do so do it this quickly because it can be annoying for people sitting behind you seeing a cell phone or iPad up in the air so anyway the whole thing's gonna be available and on the vault afterwards before we start this is the the content will be seen today during the presentation I'll start with an introduction part with some of the challenge we wanted to tackle and then the objective for the pipeline itself and we'll take a look at the procedural tools that we develop for it then all those tools are used from a user point of view and after this we'll dive a bit more in depth into the pipeline itself and we'll also look at all the the cliff and the the biome tool or working on the David unfortunately I didn't have enough time in my presentation to go in detail in every

tool that we made but we'll see those two at least and then a quick section things that we changed along the way and finally a conclusion part so what was our initial challenge we wanted to tackle first we we had a tearing that was changing constantly this is the Far Cry 5 tear in evolution over two years and a half how do we manage or content when there are so many iteration like this will have the same evolution here and a smaller portion of the terrain we have a forest that has been applied manually and its distribution is initially coherent with the terraforming and however over times it becomes really in current as the the tearing change so how do we solve issue like that it would be for sure tedious work to repaint the forest every time the tearing change also if this was done and if it was

managed by several different users how consistent with the result be should we like terraforming early in the project to avoid problem like this I don't think so because this all iterative process is vital to the game quality Indian it is with those challenge in mind that we built a procedural pipeline or first objective was to develop a macro management tool to fill up the world with natural-looking content then this content needs to be consistent with the terrain topology as well here we have the same example as before however this time the forest distribution is that that adapting to the latarian shape so as we can see the the result remains coherent with any of the terraforming change then the whole pipeline needs to be automated as well we used with xenian with an engine as well as a nightly

generation on-bill machine to fully refresh the world every night robell machine that will process different part of the world one by one until the food world is generated and this process ensures us that the user will have updated world data every morning then all the data Jenner is generated also needs to be deterministic which means that the generation needs to yield the same result given the same inputs so by using different build machines whether we bake the left or right chunk of the terrain here the middle part will always give the same result and this is quite important because for the navmesh generation among other things well we have the nightly build that makes the world map per map so the junction between those maps needs to be seamless and finally your last objective we wanted the whole thing to be user-friendly just the nightly build alone is not enough to ensure user have

up-to-date data you need to be you need to be able to bake the procedural generation as they work as well so of course we have all the tools available indeed stir now let's take a quick look at the available procedural tools we develop initially our mandate was just to tackle the biome distribution develop a biome tool but we ended up extending way beyond that first we have the freshwater tool that generates lakes rivers streams and waterfalls we have a fence and power line tool we have the cleft generation and steep terrain surfaces then the biome tool that spans all the vegetation we also have a fog density map generation that is then used by the fog shader and finally we also generate the terrain there sorry the wall map terrain mesh along with the

scattering of miniature trees on top of it but I was it like using those tools from a user point of view in this next section we'll take a quick glimpse at how it is like to fill up an empty with the procedural tools so first off user will terraform the Turing using the leads to tools and then the artists can lay down a freshwater network that is achieved by using curves and spline that will be used as input for the freshwater tool for rivers we have control and on the spline for the web and once we're happy with our inputs the user just have to take the freshwater tool and this will produce water surface as a lesson Tarun texturing and scattering of some asset under water then the user can generate the cliffs the tool is simply

based on the terrain slope so you basically just have to run the generation and the desired terrain area we also have a few extra control like exclusion masks and stuff but we'll see more detail on this later then the artists can add the vegetation so this is done using the the biome painter here we just said I can apply a Maine biome to the terrain then run the procedural generation so the the main biome will naturally distribute grass and forest biome and the the recipe will also react to water proximity spawning specific asset it will also avoid spawning vegetation on the cliff erosion line that was generated by the curse tool and also in this case altitude also affect all their recipe behaves spawning

smaller and sparser trees up in altitude so this is nice but the user will certainly want to customize the location so what the artist can do is he can start by clearing out an area by painting as a biome like grass then laying down a road spline so so far our result was quite natural but that's all we will bring man's influence in the environment and override mother nature so once we're happy with their input we in this case we baked the roads and refresh the biome and now we get our updated result so the trees are cleared out of the road here and of course we there will still be some asset that will be need to be placed by hand like a house shed and vehicle here yeah the level artists are not out of

work yet based on those perhaps we need to readjust the background painting a little bit so some biome brush that we have they don't do any asset scattering but simply apply a terrain texture so the user can use that to create some kind of driveway and then we can also add back some trees and forest using the the forest biome so we can apply that here and there so once or a key happy with our setup we just refreshed the biome here and we see that our driveway is not here at any grass and we have our little patch of forest where we paint in them and now just to cover all our tools let's take a look at how we can add a fence the input for that is once again the spline so the user will send please set the defense type he wants to generate on the spline perimeter then run the generation

so here we get a nice wooden fence user can also add powerlines network in the world the electric poles will spawn on each control point of the spline in this example we had three different types of power lines that will automatically connect with one another so we added a standard line a single line and then two hours connectors we make sure to snap the splines together then we run the generation and the tool will nicely add the transformer transformer box were required and also you're the user didn't snap the splines together but the system will will do that automatically within a certain distance threshold here we have a bit of problem I've got trees in the way of the power line so easy enough to fix we just have to refresh the volume

tool to take the power line to account and the trees will be removed because the systems they can speak with one another so that that covers up the the tool from a user point of view but so just to illustrate the non-destructive workflow the the terrain can be registered at any time the user will simply have to like rebate the affected procedural tools so back there I lowered the terrain quite a bit and I didn't adding an aisle in the middle of the lake so in this case I just have to remake the the cliff and biomes and voila we all did the vegetation is updated according to the new terrain shape and we get our nice little island in the lake so now let's take a look at how this is working under

the hood we have a new Disney engine and ployment a ssin with our in-house engine and it's der dunia and what is opening doors and possibilities with this pipeline really is the available inputs and outputs the data exchange between the two and well quickly for those who don't know what zinnias it's a 3d software with a node-based workflow that allows you to develop tools that we call digital asset or HDA in short then those tools can be run in the woods in the engine API when implemented as a plug-in with a specific application like a game in stirrer some of the inputs are sent from dunya to whitney true Python script so we have the world information a well name the size of it because we didn't have only one world during production we also add the gem for testing purposes then we'll stand over the file paths also because

we can expect all pcs to have their stuff installed at the same pad and then the Turin sector which is basically just the area of the terrain we we want to generate I will see more detail about this in the moment then the splines and shapes along with metadata as geometry attribute for instance the spline would have defense-type as a as a geometry attribute then other inputs are simply extracted and Zisk and available to read from Houdini with the file path provided by dunya so we have d---eight maps the biome painter PNG is to DT rain mask we also have access to with the new geometry that might have been generated by specific procedural tools but the main input for the procedure all is is the terrain itself because the generation is is always linked to a specific area of the terrain in our case and so this is how the terrain is subdivided in indeed stir for us our

smallest granularity is a sector which is a 64 by 64 meter which means it is the smallest area a user can bake what happens when someone wants to rebec the procedural data he will select this generation area so it's either all basically everything that is loaded it's worth noting that in dunya the old world is not loaded at at all time indeed stir the world is divided into several maps and when opening the instr we we select which map we want to load then the other actions are the local map section or sectors the local being the the one located right under the camera and our last option a frustum mode where we will simply bitbake all sectors visible by the camera so now let's take a look at the data that Whitney can send back to the editor we have the entity point cloud there in textures there a night map to determine that either RGB or

grayscale and procedurally generated geometry as well as Tarun logic zones which is basically just IDs on the terrain to specify an environment preset and past process then all of these are saved as a buffer in a special format on this and that gifts time today it's to to lock the data at its own pace and that avoids sending one big buffer and that takes a lot of memory some quick specification concerning the entity point cloud it could really be anything that as a position in the edster so vegetation has said rocks collectibles decals VFX prefabs could even be animals and their favorite type of environment the points will be exported with the object ID of the asset to spawn in the instr then the exported

data is also key to the tools interconnectivity the procedural generation is sequential so the bill machine will always bake them in disorder so each tool can export data with the goal to influence the following ones so some tools only write new masks and others will read masks generated by previous ones as an easy example we have the the freshwater tool that will generate a water mask that is then used later by the biome tool to drive the spawning behavior or some some specific species so now for the cliff so the purpose of this tool is to generate clicked geometry on large terrain surfaces it acts as a detail skin on top of the terrain so this is quite a large section so just to give an overview the content we'll see in there we'll take a look at the previous deck the tool input

stratification geometry shape shading method the Terran data erosion vegetation growing surface and the the exported data bein so or pre stack well there was none on on previous part dry beside the endless rock the cliffs were only like bare terrain wasn't there back then but I was told that the world used to be designed to design to avoid having too much large cliff surface but now on for cry 5 I don't know what happened but we stood in the other world with like a large amount of cliff surface they were they were much bigger and than before and invisible from much shrewder as well so that's the reason why we decided to develop attack to improve the situation about that so this is a shot from the famous fc5 without the cliff deck and

with the new cliff geometry on top of it so how does that work in you easily we use the terrain slope attribute as a starting point and then delete all surfaces below a certain slope trestle and this becomes our cliff input geometry as an additional note doing that is interesting as well because it also serves as a visual cue for the player to know which which terrain is too steep to navigate and this is exactly the right slope threshold that we used to generate the cliffs since we start from the terrain mesh we get stretched quads on the slope as we see here so we get rid of those by doing a remission pass on the Jeremy tree to get uniform triangles and then there's an interesting phenomenon that we want to reproduce with the tool and it is the geological stratification the visible

result all lines that were formed by the the accumulation of a sentiment every rock and soil all the time to create this effect we we have a tool that sliced the input geometry into strata chance each strata as a random thickness it will assign a strata ID to each slice which we can use to randomize the strata color as a debug view as we see here and then in the editor we also have a preset to control the strata angles the user will control them with this RGB input painted on the terrain then Whitney will use that to to drive the stratification and gold parameter in as a result we get the strata lines strata slice in three different angles here the angles that we chose were some that bested the terraforming of our world however just just like this the strata lines are a bit too perfect and unnatural so to break to break them up and bring some

chaos we do something else so we start by generating a noise that will be used to split the input mesh into two groups the noise is generated on a lower resolution mesh and and transferred to the iris one and we do that to get larger and block your patterns using that noise we split the cliff surface into two group like this and then we run the stratification to on both groups with just a different seed value so that breaks up the destroyed lines after this the geometry is ready to get some volume so each strata is extruded at various thickness and this place using a combination of a displacement map then at this point the geometry still fairly arrests so we do have a sensation pass and we reduce the triangle count and

before export the geometry is also split into several individual mesh and the that is with one mesh per sector to be more precise and that's for better loading and stream streaming and game so all these different different colors represent a different sector now that we actually don't cut the the geometry at this at the step each triangle already knows in which sector it resides so we just have to use that to split the mesh apart so that that covers that for the cliff mesh part but the cliff tool does other cool things as well and before going further I kinda need to explain how the cliffs are shaded in game we use what we call the tearing shader some of you might have seen the talk of my colleague Jeremy more yesterday on the Terran tech we developed for sd5 if not and if you're interested you know you should look it up what's in it once it's

available on the vault anyway what the shader does is that it will render the object in the same path as the terrain so the cliff will use the same texture as the terrain directly under needed so if we have grass underneath the cliff will be partially texture with grass too as we see here so to avoid this we need to make sure that the cliff well we have cliff there in texture right under the extruded cliff mesh as well so for that reason we use the final extruded cliff mesh and we recast down some cliff attribute back to the terrain so we do a cliff mask and Astra we transferred it to the strata attribute as well from the strata attribute we just transferred we first generate a color layer that will create a macro tent variation in the world as an added additional note here

this color could have been on the mesh vertex color but we didn't do that for two reasons first because we want to maintain the tent on the terrain you even in far distance when the the cliff smash is actually unloaded because because of the the kill distance and also at the same time we wanted to really reduce the cliff mesh memory cost as much as possible plus we already have the color modulation layer in our terrain layer intact on which we merge this color then from the cliff mask we transferred to the terrain we further extend the cliff by running a flow simulation the points scattered on the cliff surface will flow down the slope to create this erosion effect and it's worth noting that the original strata color is retain on the erosion area as well then using that erosion data we also scatter crumble rock entities on the terrain those will be exported later as a point cloud to the it stir then

from our mask we generate the terrain texture IDs we we also use a noise to mix two different cliff texture and we do that just basically just to break out the texture tiling another feature of the tool is that we can also scan a scatter of vegetation on top of the viable cliff surface in order to find which surfaces viable we isolate the polygons facing upwards and simply check if it's clear above or not using a recast and quick result in game of those cliff edge with vegetation and at the end this is the data exported to date story we we send a cliff jump tree along with our collisions and hence the point clouds for the rocks and vegetation we Center in texture IDs of color and a

cliff mask and this is some interesting procedure octave shots from from the game okay for the biome tool now so the goal of this guy's - is to populate the world with biological content bear with me here this is probably the largest section of the of the presentation there's a lot of feature to cover mainly because there's so much adulty in nature and so many various phenomena to reproduce so the first step that we do in the in the biome tools is we generate the terrain from the right map so easy enough then from the topology we generate a biotic data which are basically the physical features of the land so we generate an occlusion flow map the slope a curvature illumination or sometimes it's called a slope aspect then others

like the altitude let stood and also a wind vector map so something important to note here is that all those tearing attributes will become the the pillars - most biome recipes then after that we impart other - to the data that do new extracted for forests on disk so we have the the bottom painting done by the users and procedural data generated by previous tools such as the freshwater road fans power line and cliff masks and then we're ready to roll with the fun stuff so first off something really cool that the biome tool does is that it does not simply spawn forests where the users painted for us we have what we called main biomes and sub biomes the main biomes are covering probably well the most part of the world probably like 75 to 85% and the main biomes will automatically process where the sub

biome should be based on the abiotic terrain data that we just saw and it is this part of the tool that gives us this natural looking macro the detail in the in the biome distribution so and then this meant by on also process other fancy things like replacing forests with your ass man where the user placed power lines after this process we have our sub by own recipes they're processed and chain and each of them contains all the ingredients that define this recipe so trees saplings bush grass etc and at the course the biome recipe all of these note are the generate Terran entities HDA and this guy among other things do the vegetation scattering and can modify and create terran attributes but a key part of it is that it defines viability

for each species so each species is fighting for its ground to grow and drive on the viability is define by setting up a favorite Terran attribute for each species and the species that accumulates the the most viability at a location will win over the others as an example we have a species a here that likes to grow on this range of occlusion terrain that as we see on the button level with a power of one and then we have a species B that likes to grow on this range of flow map with the power of two so the species B will win over species a where the flow map value is high enough but to determine the winner there is also a ridges that is involved here we have a species species as our viability than the

species be so we see the radius in color on the terrain if species being ends up inside species a radius will be removed like the one in red that we see here so this is nice we keep at this our ball spacing between our tree like this however what if you would we would like to allow bushes or grass to spawn underneath those trees we wouldn't want those to be removed so for this we also have the priority with the priority ranges the priority will actually be evaluated first and then if two entities priority are equal then we will then evaluate the viability instead the previous example has been extended ear however now or green tree priority radius is smaller than its viability ridges all our tree are set with priority of ten and we added the bush species in yellow with a priority of zero the the yellow bush will be allowed to grow closer to the trunk of the green

tree now which has a small priority raises of two but if we want to we can still prevent them to grow to close from the blue tree which this one has a larger priority radius of ten so we saw that viability as a big role and vegetation distribution so now let's take a look at some natural phenomena here on the left you can observe that we don't have any vegetation growing on the flow lines in the steep slopes and on the right we also have almost no vegetation growing on the South face of those mountains so well to mimic similar phenomena we kinda need to be able to combine different abiotic tearing data together it is by mixing various tearing attribute that we can create very specific patterns or a species distribution and accumulate a fluctuating viability as well so here as an example I I multiply clusion by a range of

altitude and in parallel we can also mix another set multiplying flow flow map with a different range of altitude and then we can add the result of those two sets together and additionally to create extra chaos we can combine noises with the Turing data on the right ear we well this is a standard problem noise with control option for scale and a set however it can also be warped by the Turing normal to create more interesting patterns as we see here then we will we would also bring in various exclusion masks generated by previous tool such as a fresh water road or cliff mask and apply that on top as well and for this example that's what we would use as viability for the species so as we see

combining turn that noise at the core of the biome generation were true however the viability also drives other effects like the species size in previous production the the asset size was just well it was limited but it was also just randomly selected it had no car egrants with the terrain or the likeliness for species to grow in a specific area now our tool can handle multiple styles for the same species if we take a step back here there's there's various thing that can affect the tree size we'll have like a small and young trees they will tend to be at the edge of a forest and at all in all trees will be more present at the core of a large forest patch there's also other factors that that come in play like the altitude that can affect the tree size so a quick example of that here so how do we manage our tree size

in the tool well we link the asset size selection to our viability so here we have a gradient viability on the terrain and we start with one conifer of 50 meters and if we add the second one 40 meters 30 meters and so on we see that each size will be positioned to the proper proper veloute range on the terrain and this will give us a nice tapering effect at the border bar for us however we have one more problem and it is some kind of a staircase effect so to get rid of that we we can allow each size to scale up or down by a certain percentage just to bridge bridge the gap between sizes and then the next additional option we can also play with a random scale if we don't want to mess with our viability too much to create more chaos and of

course two trees of the same size are not necessarily alight so we can have several variation and the example here we have a living tree in the deck a dead tree variation and were just playing with the probability of the dead tree to shift the violence between the two let's take a step back once more and look at another phenomenon the forest canopy and the ecological succession as I mentioned earlier young trees will tend to be growing at the edge of our s and then the old ones will be deeper inside but there's also a younger grow possible inside the forest so how can we achieve that we said of the viability can affect the tree size selection however depending on the tearing data that we use we don't always end up with a smooth gradient on our viability which can result of having our tallest tree at the edge of the forest for instance so

for this reason we added the the H parameter which is basically just assigned distance field generated from our viability and well we have various options to control how it will affect the size distribution we can either add multiply or interpolate with the viability so we we can easily tweak the level of influence it as like this and by adjusting the age maximum distance on the Sanga Stansfield this gives us a nice control and all the the the border of our forest will be and as a bonus side effect by using a ramp on the edge we can also profile the shape of our forest there's there's another important thing we need to control and that is the scattering density if our density is

uniform well we might have a good spacing between our small trees however we can we will have a really bad overlap on our large ones and also the number of asset we spawn is a matter of performance of course the small trees they don't have the same GPU cast as the large one so it is very important to keep a tight control on that a sample ramp ear will give us that control by default the density ramp is based on the size of our species however we also have other options to base it on the edge or viability if we if we have a species that don't not have like several size variation then we can also mix in teheran attributes like elimination or slope aspect to affect the density of spc so an example of that year and example was the same thing in game

jumping to another feature now some of the biomes we wanted to recreate add a lot of color variation here the grass and bush color is varying a lot in this reference picture so to do that without duplicating our asset texture or material we have the option to tense the the scattered instance individually so the input value that drives the gradient color ramp can be either viable te age or a combination of any terrain data and here in this example were using the the water sign the sign distance field data - just to give a rust attends to the grass growing in our wetland biome and well the the result in gain here that that's where this rust color variation comes from the scatter density also needs to have their own specific rotation by default they have their forward access

oriented towards that are in slope and this is interesting because it it allows us to do things like having this u.s. asset that we see on the left ear leaning toward the water and that's because of the slope so either it's on well it's on either side of the if it was on either side of the the river here the grass would be always being toward the water another example is a prebend that tree trunks that that rose in slow unfortunately we didn't do that in our game but I wish we did it's kind of cool but we could ask the system would allow it we did other cool things Oliver and well not sure if any player will notice that in the game if I was not mentioning it but this this grass acid that we see here well it is oriented on the wind vector map so this is another orientation option that we have the wind

map is just based on overall wind direction but it is fluctuating slightly based on the terrain there in shape as well so it might be a bit hard to see here but well when blowing against the hill it will tend to flow around it a little bit then now just to finish with the rotation options in all cases the acid can be horizontal or aligned to the latarian slope that's controllable with a percentage and we have a rotation jitter on all axis where for we use that a lot on oral or vegetation essen now we we have covered most of the parameters to control the entity generation itself however the biomes are not limited to the asset placement only because the position of asset can also have an effect on some of the terrain properties and it can also have an effect on their surrounding asset as well here in this picture I see four different things that

the present of those trees and influence visually in the surrounding first the terrain is covered with pine needles so the tearing texture is affected we have a slight tear in elevation around the trunk of those trees so the the Terry Knight map would be affected as well we have pine cones and that branch on the ground around the tree and for sure the the shadow provided by the by the trees might encourage or prevent the growth of other species so to do that in our system after the entities distribution is done we have the option of transferring some data back to the terrain from the scatter densities and we use we use this process for four different things terrain deformation basically just affecting the ID map tearing textures tearing data output which is a mass that can be used in following species and then a tear in

color this is just a team that gets mix in with our order in texture and all four all four of these are independent from one another let's start with a certain example with the Terran deformation we can generate a mask from the scatter densities position and then optionally we can combine it with the the Apple available Terran data as we saw earlier for example I might want to mask out any any terrain deformation close to the road so I'd be using the the road mask in this case and once we have our mass we said the the displacement ID for the terrain here it will be raised by one meter and this will deform the terrain accordingly so this is the for the deformation we get from a user point of view and I think this is a really interesting feature to have because most of the time in games the trees are just taking the ground they they they don't have any effect on

the terrain around them but in nature like the tree roots will be lifting the ground a little bit and also by holding the soil together that would that will limit the erosion and then create a shape like this in the forest now let's take a closer look at those tree roots we will want them to blend seamlessly with the terrain to do that we can generate a matching pair in texture underneath once again we generate a mask from the scattered entities and we set the number of texture we want to generate from this mask select the desired texture from the world roll down menu which will actually fetch the available tearing texture from these two via Python and then or mask grayscale value will determine which texture is applied where and well we said so we set the second texture here

to appear where the mask value is 0.5 and above once the texture IDs are some - its - this is the same kind of result we would get and now our roots are the ending well with the terrain underneath next perhaps we want to add the extra ingredients around those trees so what we can do is generate a new terrain that attribute from our trees but this time this mask will be actually stored on the terrain to be reused as viable tea by the following species once we're happy with our mask setup or ready to go over to our next node forest rocks in this case so we define this asset viability by using the Ponderosa mask we produce with the previous node and as a result we get those asset scattered around those trees and so that the position of those will always be linked to the position of the other species so that's that's quite convenient if we go back to

our Panda house a tree node we can also output the edge data for a species and which will allow us to for instance to spawn a new ingredients just at the edge of the forest for instance or if you guys remember as we observed earlier we can also use that to create the young effect for the ecological succession then we would just keep adding further ingredients to build a full recipe like this and on the 1:1 map of one kilometer square in the game that's over 600,000 and steez generated by the volume tool and for SC 5 we add about 70 maps in total like this in the world one last feature this is Google Earth picture and the phenomenon were interested in here is the the Terran humidity the color

variation between the dry brown and lush green so as we suffer the the cliff's rata color we in the biome tool we also generate their in their intent error that is that will get mixed with our organic Terran texture distribution and this gives us further variation and at the same time we limit the amount of texture tearing texture we need to use in-game note that the the neutral color for this color layer is an average gray which allows us to either the darken or lighten the terran texture and you're just zoomed out view of the terrain in game just to see the texture and color variation generated from the biomes it's also worth noting that all the Terran texture and procedure alton can also transpire through the grass and that is because the grass shader will will pick

up the tear in color and well the amount of calorie picks up is also controllable with a mask of the grass asset itself so it can be either like picking up on red percent of the terran color or just as a gradient or onion specific part of the plant and at the end this is the the data exported to the in store density point clouds there in text released arena Taron color and also a forest mask that will be used later by the fog and the wall map tool so in this section I want to share some stuff that we changed along the way while things don't always go as planned some ideas might be fine on paper but it might not be so good in the end so a first example of that is for our biome painter initially the plan was to support like a gradient painting

so that we could blend blend the biomes together it is it was supported in the you Disney tool because the viability system would would kick in to merge all the all the biomes together but from an editor and a user point of view it was not such a good idea turns out we we had so much biomes that could be mixed together it would have been a nightmare to debug so on the left here we only have three biomes and it it gets really difficult to tell all the surface on which the red biome is painted for instance we we could have a mode to visualize one biome at a time for sure but it was it was great for the user to be able to see all of them painted on the terrain at the same time and really in the end that trust fade blending was not needed at all to get good result so here on the right even though it looks strange this is the input format that we use for the biome painter another example is the Terran

abiotic data generation we saw that at the beginning of the the biome told earlier but it is actually using multiple tools so at first we had a button a separate button - to generate that and cache it on disk and then if there was a terraforming change the user had to remake it before generating other tools like the biome tool but what happened is that people got confused about when it was required to bake it and as a result it was not always rebec after terraforming change which led to different result and the real final result given by the bill machines which was actually remaking everything properly so well they're the reason why we separated this process initially was was really just to say that the tough time when making these other tools but in the end this was an extra step that

brought complexity for the user and it was not worth it so I will simply decided to incorporate that inside the tool that that required that data that we were sure it's always it's always up to date finally a conclusion some lessons learned with great power comes great responsibility I always wanted to say that procedural tool can generate a lot of data this gives us a great control over performance but also for gameplay and art and s5 the art direction they wanted like crazy dance for us however from a gameplay well gameplay wise it was not interesting at all because the AI and the larger animals they could not navigate in the forest so on our side with control over the the procedural vegetation distribution we have to make the right call and it was just to ship

with the most density without impacting gameplay another lesson make sure to design elegant tools that opens up possibilities a good example of that is or biome tool it's it is a system that allows so much possibility honestly I think we barely scratch the surface of what we can do with that and that's good then make sure to keep keep things simple well any new systems are often often open engineered and that is because we figured out each of them and each element of that system at first but once everything is in place we have a clear view of the whole thing and we can most likely do a clean a pass and a simpler more design can emerge then make sure to listen and observe your your user they might prefer to have manual control over an automated process and a good example of that is or freshwater tool the rivers

they could actually carve their their own riverbed automatically but but it was optional but actually during production like almost no level artists use that at all they just prefer to do the terraforming manually and then place the river so which leads me to my next point like make sure to be flexible initial plans are not always the best and then make sure to keep a good balance between control and automation too much automation and really things can get out of control if you have like many systems that are independent from one another and then just changing one thing here is actually affecting something else elsewhere that might be really bad and on the other hand too much manual control and things can became become really time-consuming and difficult to manage and after all this is why we're building procedural tools just finish I want to give credit to all

these people for their small or big contribution to the pipeline and well I hope you enjoyed the presentation if you have friends friends that they didn't get the chance to get in or you think that they might be interested to see the representation now I'll be giving the talk again tonight at the Ubisoft lounge from 5:00 to 6:00 p.m. so thank you everyone [Applause] any questions don't be shy well hello okay there you go can you give us an idea of the resolution at which you're storing a lot of these masts and terrain data like Iike topes

for example yes are the pilings so the resolution of the mass is actually one pixel per meter on the terrain right sorry that's about resolution of the train itself yes exactly the the terrain ID maps is actually 50 centimeters okay testing incredible time stuff thank you I had a question about the point clouds yes does that what are you writing out of Houdini's is like a big file full of it's like vector data and then the engine reads it in yeah well actually we have like a proprietary format that we call game X and it is its use across different engine at youssef and yeah it's just a file format that which we will start the the points and the position and all the the attributes required for for the engine that's going

all right thanks hey good question about the users I was wondering if you had a system to compensate for the errors made by the users in terms of the terrain let's say for example you you're talking about those squares of like one killed in a war on kill Manor yeah what if I sculpt the edge there and I don't have the other to shake a place and there's some source control what if I change the edge of this terrain and the other edge doesn't match anymore or I put a river there or like and it doesn't match like every field that afterwards or what very clever question well first and to do discuss actually the the tool in the editor will well prevent you to paint too close to the edge just to avoid the getting scenes but let's see if you would just hold one map and then raise the terrain and then note that the next map and then you would have a really like they didn't gap between the two but actually you're not allowed to paint like to the very edge of the okay thank you hey can you give us an

idea about how many people and about how long all this work took for to develop the tools yeah good question well we were like one-and-a-half person over like two and a half years then at some point that the tools were fairly stable but we just kept adding new tools over time too because some some were already quite stable and yeah cool thanks oh hello thank you for your talk I have two questions one is about the cliff mm-hmm how did you prepare like the mess did you like have a module cliffs and then you input them into make the cliff tool or something like how do you make the

cliff edge tourism oh yeah well as we saw is just we just start from the the tear in mesh and then just modify the geometry from there so at first the the input mesh is just is just the terrain and we delete some polygons of it and just keep the actual slope range that we want and then by doing like a drunk geometry operation on that extrusion displacement this becomes our actual mesh when working with height fields it intends to like rivers go from up to down but then yeah time is you like when you do the wrong path it goes down to up yeah you um control that it was up to the user to make sure they they set up the terrain properly to have

like the the water flow down in a in a proper way it was not a big deal if well we made sure that this didn't happen but if it was the case it was it wouldn't break anything it was just it would it would be weird on the on the flow map itself then you would have a point where the water would flow down flowing down into different their opposite direction but yeah we didn't have any like check system to to avoid this with just the like the level artists they just did a good knock terraforming job and make sure it was always logical that nothing was really wrong yeah and we also had the the debug tools to to check the actual well to display did the flow map direction so it would display arrows on the water so it was easy to

visualize I can if it was flowing properly or not thank you welcome you spoke a little bit about avoiding lucky and anything in the terrain so that the iterative process can continue um just wondering a little bit about artificial things like roads and buildings would you bring why certain types of terrain to be placed mm-hmm um how do you deal with that workflow and when we need to lock in the terrain before adding those yeah well any any terraforming with while altering was done indeed Styrian yes the the roads would also affect the terraforming but that was like done indeed stir as well had because we have the the road tool that it's actually uh needs to to but then for the procedural the like the procedural tool would receive all the terraforming what the right map would be extracted with the the terraforming of the road and any

back like a flattening for a location to set up a house and stuff like that soon so if you're planning a village it the planning would be completely different depending on the terrain location and would you plan the village after making sure everything is finalized there sorry would you plan a village layout after you make sure that everything is finalized I'm not sure understand the person can their chat afterward okay more in detail okay alright well thanks everyone [Music] [Applause]