Wakes, Explosions and Lighting: Interactive Water Simulation in Atlas

[Music] my name is tim chavlakov i'm senior developer technology engineer at nvidia and let mark introduce himself and introduce us to atlas my name is mark mahelich i'm a engine programmer at grapeshot games and we're currently working on a game called atlas and uh what is atlas uh alice is a pirate fantasy mmo but it's made by the people who made uh arc survival evolved so there's some survival aspect to it as well we're currently in early access you can get it on steam and i have a little video here it's an intro so we can play

[Music] so as you can see um a lot of your time in the game is spent uh on the ocean and looking at the the water surface so we really wanted to get the most realistic possible uh water simulation and rendering um so we enlisted the help of nvidia and uh we're you know the game is still under development but we feel pretty proud of what we've achieved so far a way to get a lot of things right um first we obviously had to simulate the ocean state we have to handle the physics of buoyancy uh of course rendering the

ocean surface and then other interactive features such as wakes explosions and detail particles so those four components are basically what the agenda is for today for this talk and i'll let tim start by talking about simulation thank you so in order to keep you guys not distracted by the rendering this screenshot and the following screenshots are just black and white so it's just the fresnel term rendered so you can focus on the displacements we start with the simulation of the sea states we use vanilla spectrum-based approach evolved from famous tessendorf's paper called simulating ocean water and essentially what you do there is you generate the wind wave spectrum you evolve it in a frequency space and you transform the data to the spatial domain using reverse fft steps then you just rinse and repeat for any

moment in time you can just calculate the c state and frequency space and convert it to the spatial domain this simulation has a bunch of good properties first of all it only depends on the time and the spectrum state so at any moment in time you can know uh the displacements of the ocean surface it is very good for us because we are working on mmo game so the simulation has to be similar for the server and all the clients so we can run the simulation independently using those uh properties of the simulation and what is more important is that all the servers on the grid can simulate the same c state so the transitioning of the player from the server to server uh can be uh seamless uh why is that important because the uh online portion of the game the servers are forming the grid and each server

simulates a square of this much square thousands of square kilometers of the ocean with all the uh players and islands and actions happening and when the player moves to another server this transition happens seamlessly so the results are tiles with periodic displacement data and the results can cover infinite planes of the ocean surface which is also good for us but there are some issues with this simulation it's not good enough for us and there are a few reasons for this first of all it uses fft and ffts as you know are expensive so if we use larger 50s we sacrifice the performance if you use smaller ffts then we either see the repeats that you can see on the right portion of the screenshot or if we increase the world space area that our tiles are covering we don't see enough details and this is clearly visible on the left portion of the screenshot

to fix that we use a advanced approach so what we do is we still use small ffts but we do split our spectra to several frequency frequency bands we decided to go with four frequency bands and each frequency band for us stores the wavelengths of a certain range so to simulate the ocean surface in this case we split our spectrum we evolved the frequency band in the frequency space independently and then we just convert the simulation results back to the spatial domain using the reverse of inverse fft steps then we post process the results to get an isotropic brdf we're going to talk about this a little later and then we just combine all the uh displacements from all the frequency bands to get nice displaced ocean surface this

is how it looks from left to right are the frequency band from the lowest resolution one number zero then goes number one then goes number three and then on the right portion you see all the frequency bands combined and if you are attentive enough on a frequency band number three you will see that there are some repeats and it's kind of weird isn't it we tried to solve the problem and we introduced the very same problem it's not something good right and if you take a look at the uvs for those frequency bands in a world space which are marked white for the most uh higher detailed frequency band then goes uh blue then goes green then then goes red you will see that the white cascades are pretty small and obviously they will show some repeats but in fact they not they are not showing the repeats because the lower frequency bands the

larger waves do distort or displace those small cascades small frequency bands enough so that we just don't see the repeats at all another thing that we didn't like about the vanilla tesendorf's approach was the phillips spectrum it is the spectrum that defines or describes the c state for the fully evolved c which means the wind was blowing over the ocean surface for infinite amount of time so that the waves fully evolved and what would happen if you try to apply this phillips spectrum formula for a puddle with the wind blowing with like 100 miles an hour speed over it you will get something like 20 meters wave or something like that it's obviously not something that you would see in nature to fix that we just use better spectras we use dual john swap spectra and john

swap is a d abbreviated as joint north c wave project a bunch of scientists just gathered together and measured the wave displacements in the north seas in northern northern atlantic using the boys satellite data and they generate the formulas for us that we can use so we use dual john swap spectra why dual because it will allow us uh to simulate uh like two spectras for the cost of one so the first spectre will simulate uh for instance if we want to simulate the small medium local waves generated by the local winds and the large decaying waves generated by the storms like 100 miles away long time ago uh these uh long waves will be generated uh will be simulated by the second drone swap and this is how it looks it's uh uh artificially emphasized so that the long waves are too

long and the short waves the the local waves are too too small but you can see uh what i'm talking about now these john swapp spectras uh allow a lot of artistic control compared to phillips spectrum so it's wind speed wind direction wind fetch this is the parameter that defines how long the wind was blowing over the ocean surface or how big was the area that the wind was blowing over then the spectrum peaking that tells us how dominant are the dominant wavelengths so that the uh the peaks the wave tops are more pronounced more pronounced than the directional distribution which which tells us how chaotic are the uh waves so how big is the portion of the waves that do follow the wind direction then if you want you can override the amplitude you're going to be not physically correct but for like artistic purposes you can emphasize the heights of the waves or dampen them down if you want and then low pass filter it's pretty

useful for uh swells you don't want the swells to have the smaller wavelengths and mark we'll talk about the beyonce buoyancy and not beyonce so we've covered how you generate the ocean displacement data um so now i'll cover how um we can use that data to have objects in the ocean interact with the ocean simulation and that interaction is basically physics of buoyancy now you may be familiar with our community's principle basically it states that body and fluid experiences a buoyant force equal to the weight of the fluid it displaces and the direction of that force is opposite to the direction of gravity now if you notice in this formula here the main unknown that we have here is displaced

volume or v and with our complex ocean simulation calculating that can be kind of complicated so how might you approach it um you might take a bunch of uh you might take a bunch of discrete um displacement samples and uh along the whole of the boat and um oops excuse me then you can use each sample assume it represents a cross-section of displaced volume and apply it back to the previous formula to generate force the corresponding buoyancy for each sample can then be applied back to the rigid body simulation uh your game's in your game engine's rigid body uh physics simulation can handle all that um so here's what that kind of looks like you have um oh it's going really fast so you have your discrete samples here discrete displacement samples and

each one of these red arrows represents a independent force acting on the whole of the boat and obviously this is a 2d representation we would actually have an array of 3d forces and then here's what that looks like in code it's um it's actually fairly straightforward as you can see you would basically call it this is happening on the cpu you call this in your your physics update loop each tick for each boat in the um each relevant boat i guess let's just say um so we had we came across some issues when uh just using pure physics simulation um one of the main things we found was that the discrete wave height samples can be quite noisy um and this is because we're we're just sampling uh instantaneous uh instantaneous location in it at an instantaneous time so um we're not you know our

wave simulation doesn't allow us to integrate over time or space so ultimately we're not able to take enough samples fast enough especially for uh server simulation which can run quite slow uh the other issue we came across is just not being able to control the dynamics very well and we found that using just physics alone creates some instability and latency and then finally maybe most importantly we have quite epic waves but we don't necessarily want our players playing the game to get c6 so we decided against doing a pure physics simulation instead our solution we still use the indiscreet wavefight samples but instead we feed those samples into a plane fitting algorithm i'm not going to cover how to do a plane fitting algorithm but i just want to

call out that david eberle's geometric tools has a symmetric eigen solver that might be a good place to start and that's under the boost license so so once once your plane fitting algorithm outputs um a target plane uh you can use this as your your targ uh your target transform for your ship and that's what we do with each frame um and we also apply a simple spring as a filter to help filter the noise and also mimic the physics of buoyancy and inertia but before we want to before we move on i just wanted to um call out something about one of these previous slides here um this function get water height at point which is something that we do quite a bit in the game both in on the cpu and the gpu it's actually not as simple as it seems because of the way that we

represent the ocean displacement it's represented as a 3d displacement relative to imaginary plane and so we actually are just looking for the height at a specific world location and one way we could potentially achieve that is to have an intermediary step where we output a height map in world space but we decided to just use a iterative approximation instead and here's what that looks like in code this is basically newton's approximation method we're just doing multiple iterations on this and it should converge over time or over a number of iterations you could potentially uh stop when subsequent sample locations are within a specific 2d distance but for optimization reasons sometimes it's faster just to do a fixed number of iterations now i hand back to tim to talk about actual rendering

thank you so we figured out the physics the buoyancy beyonce buoyancy and uh [Music] physics of the rigid bodies floating in the water let's get to the rendering so we want the ocean to be rendered as beautiful as possible this photo is actually not a photo it's the our rendering and the screenshot is taken with the camera located at roughly like half a mile above the ocean surface so by the way even here you can see the large swell waves and shorter like the local wind waves to get to this level of realism we need to uh treat our c as a micro facet surface this is schematic drawing i'm not very good at drawing so my micro facets are just all around they do not touch each other but you get the idea right uh this is another nice screenshot which is not correct at

all because those micro facets are not microphone sets but you see those as the facets so let's start with rendering equation the light that comes to our eye consists of three portions three parts first is scattered light the light that uh came through the uh water body and refracted towards our eyes then the specular component from the sun and then the environment mapping component the light that was uh that came from skies and surrounding geometry and whatnot and this fr function that we need to integrate is the brdf everybody familiar with the pbr should know what is that right and microphone said brdf model in our case and this is the vanilla case is it depends on the fresnel reflectance we're going to talk about this a little later than the normal distribution function d and the masking function it's pretty

important we're going to talk about this later as well all these formulas look scary but you shouldn't focus on those you should focus on what those mean and i'm going to explain this stuff so our micro facet surface is defined by ndf normal distribution function uh it's the function that shows how our norm how our normal normals are distributed within the pixel and this little egg-shaped thing on the bottom is the distribution of the normals and in this case for instance the the vertical normals will be most likely i mean you will most likely see the normals oriented vertically and the normals uh pointing at other directions have less chance to appear and if you unwrap this x-shaped thing to a uh 2d plane where our angle is on x-axis you will see

the familiar bell-shaped curve now the ndf normal distribution function for us we use background distribution is defined as a function of pdf probability density function these formulas are quite scary these are these are here just for the reference but they mean that our pdf defines our surface roughnesses along the x and y axis because we are rendering the waves and we need to use the an isotropic uh prgf right so uh what's important and this is the thing that you probably should focus on on is that we use uh our surface moments to calculate the pdf just like uh they do in leather paper leather is linear efficient anti-aliased displacement and reflectance mapping it's pretty cool paper you should read that so using the moments allows us to

calculate those variances uh just from the moment and uh the moments the first and second order moments uh can be stored in texture and what is cool about those moments is that they do survive the linear operators and those linear operators are the mipmapping and texture filtering and that that is something that we will gladly use right so how do we calculate the moments after we get our displacements using the reverse fft step we calculate those first order moments which are essentially the gradients the slopes of the ocean surface and the second order moments which are essentially the squares of the first order moments the squares of those slopes and then the covariance which is just cross multiplication of those and we just store this in textures and this is how it looks schematically we get our displacement textures in the

first texture array which has four slices because we have four uh frequency band we store those displacements in rgb channels of those slices then we using the compute shaders we get the first order moment and store them to another texture array and then the second order moments are stored in the third texture array and the three channels we use in those textures we use to store the foam intensity displacement jacobian and other things i'll talk about those a little later now since we have multiple frequency band we need to carefully and mathematically correctly combine all those moments the first order moments the slopes or the gradients are very easy to combine you just sum those up and the second order moments just a little more complicated to combine okay we figured out the moments we

combined all those moments we got our uh roughness values now we can calculate the specular reflection we are very lucky the specular reflection integral over the hemisphere is just analytic right because we only have one light source it's the sun and uh the integral is just analytic we take this pdf we take in account those masking and shadowing factors that we are going to talk about a little later and we calculate this scary formula this is how the specular reflection looks with the simple uh calculation just vanilla uh reflect uh specular and with pbr on the right and the picture on the right is obviously much much much more correct and what is also good about it is just the uh is that the uh

specular does not uh sparkle it's not noisy it's anti-aliased and it's smooth at any distance to the ocean surface now the environment reflection it is a little more complicated because we don't have a single light source we have entire environment map so in order to calculate it we just sum up few samples just like in this liader paper we use a set of samples which are effectively the cones which is simple our environment maps with the cones and i'm not going to go into much details about it it's all explained in the leather paper it's not the point of this presentation the thing is that we cannot afford full full-blown calculation for those samples for instance we it's just too expensive we do not calculate the fresnel term or masculine shadowing for each of those environment samples we're going to talk

about the fresnel and masking shadowing terms just a little later and this is how it looks on the left you see the simple reflection is just one sample per point and it's pretty noisy and what is more annoying is that the as the normals do degrade to pure vertical distance your reflection is just a simple planar reflection at distance it just doesn't look right and on the right uh portion of the screenshot you see nice environment reflection that does use the statistical properties of the surface now the masking and shadowing we are rendering the waves and it's quite obvious that at the distance or at the grazing angles the waves will obstruct each other right and this schematic drawing shows this

with waves obstructing uh obstructing each other right did i say it right uh you don't have uh as much light as you would think at the distance and lucky for us the approximation for this for rb or df exists so we just use this walters approximation for us this scary formula in blue on this picture and what it actually gives us uh as i just told uh on the distance you will see that the specular light from the sun gets too bright on the left portion of the screenshot and it's just about right on the right portion of the screenshot and if it wasn't clear enough let's just decrease the intensity of our sun and pretend that it's moon so you will see that on the left is just incorrect it's too bright and on the right it's looking much much better now the fresnel reflectance again waves do obstruct each other so the

distribution of normals that you would see that uh that will uh combine and uh project to your pixel will be just tilted so you will see not that the normal distribution is not uh what did i say so it's not uh ever it does not average to vertical anymore and again we are lucky the approximation for this microphone set brdf fresnel does exist so we just use it in our case and this is how it looks this on the left portion you see the planar uh fresnel that does not use the micro facet properties of the surface and as you can see as the normals do degrade to a vertical at distance the fresnel term will degrade to just one which is obviously not correct and on the right portion of the screenshot you will see the fresnel with the microfacet properties

it's obviously more much more uh closer to nature now the last but not least scattered light it's hard calculating scattered light in real time is too time consuming you have to kind of trace to the water body it's too much for the real time we cannot afford that so we just came up with a bunch of hugs and fudge numbers this formula looks scary but let me just explain it step by step so what we do there is we just multiply the chances this uh blue portion on the left just tells us that the chance to see the scattered light is higher as we get closer to the wave top so the higher are the waves the and the closer we are to the wave tops the chance to see the scattered light is uh higher then the next one the blue portion shows us that the chance to get the scattered light to see

the scattered light is higher when we look at the oceans that are located between us and the source of light the sun in our case then this third the red portion shows us that uh the chance to see scattered light is much higher if we see just straight at the ocean surface uh if the normal of the ocean surface coincides with our view vector and there is a couple uh the blue and the yellow terms these are just uh fudge factors to take in account the bubbles that are spread in the in the water and uh the shallow uh scatter we render the ocean surface just a bit like a diffuse surface so so it behaves just like a diffuse uh surface of some kind and this allows us to simulate the shallow scatter and this is how it looks luckily this

projector will allow you to see the scatter it's on the right top portion of the screenshot but the next screenshot will show it better so on the left you see no scatter at all on the right you see the sun scatter with the sun elevate with sun elevation of 5 degrees then 15 degrees and then 45 degrees it's kind of nice it's kind of cool it gives nice pictures and what's important is that scatter of the light in the water body is the thing that defines the color of the water surface the turbidity the algae what not everything that is in the water surface defines its color and scatter is what you is what defines its color now let's put it all together we did use fresnel term for our specular reflection it's all good we did not use the fresnel for environment reflection we have to use it now

and of course we have to use the fresnel for scattered light so we can sum up all those components and get to render the ocean surface one thing we forgot is the foam so we need to calculate foam color and just learn between the foam color and the ocean surface color in the foamed areas and of course we need to increase the roughness just a little for the foam uh big formed areas because the foam bubbles are bumpy to calculate the foam uh we use a pretty simple approach we calculate the display uh jacobian of the displacements jacobian sounds scary but it's just the measure of how much our water surface was squeezed or stretched and if it's squeezed then obviously we on a wave top and if we are on a wave top we inject just the float value we call it turbulent energy into the corresponding textiles and let them leave evolve over time spread over over space and just dissipate over time then of

course we need to mix or add all those foam energies from all the frequency bands and use to modulate the foam textures and this is how it looks kind of nice a couple final tweaks we did not realize that our famous formula that allows us to get the variances from the moments does not survive bilinear magnification because guess what the bilinear magnification even if it sounds kind of linear it's not a linear operator because it's bilinear so we had to detect the cases or the areas where we do magnify our textures using using ddx and ddy and then if we end up being magnifying the textures we just learn the second order moments that we get from those textures with the

second order moment that we calculate just in place just as the squares of the first order moments that we get from those textures this effectively lerps our variance to zero and makes us makes our ocean surface perfectly mirror a perfect mirror in close-ups but this is probably what we want to do right and this is how it looks on the left you see kind of weird artifacts where we are trying to magnify the micro facet surface and obviously it's something wrong to do because we are trying to use this method outside of its applicable domain and on the right you see the fix oh and plus plus we use the bicubic filtering for the normals for those gradients in close-ups so this will give us gives us smaller normals now we didn't talk about the geometry but the geometry is pretty simple we use quad tree approach there and quad 3 with lods

does have smaller density of those vertices uh in the distance so we obviously under sample our displacements with the geometry and this could lead to uh crawling geometry and like all kind of weird artifacts and to avoid that we just damped down our displacements at some distance and we empirically figured out that that 30 world spaces of a frequency band uh is good for us so after that uh distance in a world space we just start damping down those individual frequency bands displacements now mark so as we just mentioned we output uh foam and wave hat data basically to render the ocean surface and uh this works really well and um obviously you could see in that previous

slide right there that just that alone is really good at representing the turbulence of the ocean under high winds but we wanted to add some more detail by adding particles that come off the top the the white caps of the waves and one of the our our game is in unreal engine so one of the most important things that we wanted to do was allow the the artist to have some control of these particles so we wanted to work inside of unreal's cascade particle system and we're using the gpu particles there but unfortunately we can't spawn from the gpu in that system um so our solution is to create a custom emitter that emits a triangle basically projected down onto the

ocean plane uh that that represents the camera frustum and we're randomly emitting particles everywhere in that in that triangle but these particles that we emit are just tracer particles and they're not actually simulating or rendering yet and then the particles in their simulation shader will just basically look for the foam texture or the white cap features and start the regular simulation and rendering when they detect that they're inside of a white cap on top of the wave and then the normal uh cascade particle system simulation dynamics and whatnot we'll we'll kick in so here's what the modification to unreal's particle simulation shaders looks like these functions get displaced vertex that function is exactly the same function we call in our

vertex shader or domain shader when we're rendering our ocean surface and that just gives us our displacement information and then the git surface parameters that's just the exact same function we call for our pixel shader when we're rendering our uh ocean surface but we're doing it here in inside the particle simulation and we use that information just to do a simple check to see if our foam wave hats value is high enough it's larger than our threshold and then we kick off our normal particle simulation and rendering otherwise we might come back around for a few frames and keep looking and then after some time this this particle if it's not on a wave top then it will basically get killed and never be seen so that was the wavehead particles we do have other particles and um effects that interact with the ocean surface in our game currently we our cascade particle system

outputs uh particle sprites for our wakes from our ships and these are displaced by the ocean surface displacement and you can see they're kind of like decals on top of the ocean surface our explosions are just normal cascade particle systems but none of these things actually affect the ocean surface displacement itself so they are just kind of on top of the ocean surface and it's kind of misses some of the detail that way but so we're working the game is in development still so we're working on some prototypes for the future and uh tim can talk about something he's working on thank you so uh yes we need the waves uh the ocean surface to be displaced by the local objects like ships and boats and explosions and whatnot and that leads us to need to introduce some local kind of simulation for this we do

use another tessendorf's paper it's called e-wave solver it's exponential solver and the idea is quite the same as we used for the wind wave simulation so what happens there is that you have a grid of displacements and velocity potentials like i don't know like 1k by 1k uh grid of these guys then you convert those displacements and velocity potentials to the frequency domain then you evolve those guys in frequency domain using evolving operators that do depend on the size of this domain in world space and delta t then those evolved displacement and velocity potentials just get converted back to spatial domain and rendered and this is how it's supposed to look kind of nice right you see the almost perfect kelvin wig and it looks really nice

okay the solution is good and all but it's not good for us i mean we need it to be improved first of all the solution is periodic because it uses the inverse fft steps and periodicity is not something that we want to right so we introduced exponential dampening on the edges of the frequency of the simulation domain so that we don't get repeats we don't have rain gain or some weird artifacts caused by the dampening then this simulation does not simulate foam we use the very same logic we just detect the areas where the displacements are large and jacobian is enough so that we can inject some turbulent energy and we simulate the foam in those areas so overall we get to forward the 50s to convert those uh displacements and

velocity potentials to frequency space then applying those uh evolving operators is just a per component multiplication in frequency space it's pretty cheap and then for inverse ffts to get the data back to spatial domain why four because uh uh two more inverse fft are adding because we're gonna have the lateral displacements the x displacements and y displacements as well so it's two more and this is how it looks looks kind of nice but it's boring because it doesn't move let us move it move the thing okay as you can see as our virtual non-existing boat starts to accelerate it starts forming perfect kalvin wake that looks awesome and you can also see that it does respect the dispersion relation so that the longer waves move faster and the smaller waves move slower and after the boat stopped

the waves continue to propagate beyond that point so you can clearly see it's not a combination of textures or kind of hacks that we used there and it looks kind of nice and it looks really great okay the waves do propagate after beyond that point it looks really good so let's look at the thing closer this time taking uh pay more attention to those to the rendering so as we can see the foam is simulated quite nicely the foam appears on the wave tops of this wake uh it evolves it spreads over time it disappears over time and the gradients and moments from the wind wave simulation and from the local wave simulation blend together nicely so we don't see any disconnections and

whatnot so it looks good it's a perfect candidate for us to introduce in atlas right and as you can see as the boat moves and turns the kelvin wake kind of changes it's good okay so calling wakes gradients it's all nice but it's boring we're pirates here right so let's explode something something like i don't know like a barrel of gunpowder on a raft ready boom so this explosion consists of three phases the first phase you noticed that was a shock wave it's just a random displacement added on a expanding circular area then the major uh displacement area that formed the the the the major like explosion area and

then just random debris falling uh here and there okay so time to wrap up uh the simulation while uh sounding scary and too expensive and whatnot it's not expensive at all so the wind wave simulation and local waves or interactive wave simulation combined together barely take more than a millisecond on contemporary gpus and if it's too much for lower end gpus we can always scale down the sizes of those ffts lacking some details but still running the full-blown simulation that's it thank you

questions uh hi is there any hope that you will be putting all this beautiful work into a ue4 plug-in for me yes yes yes the uh plugin i'm not sure we may want to have some branch or something but yes we want the public to see it oh wow okay awesome thank you um how do you keep the uh the water from showing up inside a boat water for from showing up inside the boat mark what if you have a real or something yeah popping up and down in your waves how do you keep the you know the water from leaking in yeah we um so we use the distance fields that are generated from the static mesh of the boat hole so and then in the shader we just basically don't render whether it's the you're inside the distance field of the boat so it's pretty cheap

actually okay thanks hi how how do do simulate coast lines breaking waves and this is this part of water simulation and the second question when you combine the wakes uh how you recalculate lighting because you have a moment for wind waves and you integrate different type of waves kelvin waves and how you combine the lighting normals okay okay you should recalculate roughness make transition good question okay so i'll answer the question uh the second question first so since uh we

so the idea for us was not to use physically based rendering or the pbr or brdf for the local simulation because this thing is not going to be observed at the large distance we will only simulate like the local uh wakes and what not for ships that are like in local battle right it doesn't make sense to simulate this on the horizon so uh the second order moments are not calculated for the local waves but the first order moments the gradients do calculate right so uh to combine this stuff and be still physically correct we just combined the first order moments from the wind wave simulation just add those the first order moments from the wind simulation and the local wave simulation and the second order moments for the local simulation are just zeros this thing does not bring any more roughness that's it breaking waves mark oh yeah for the

so for the shoreline currently um we actually just kind of taper down the water to the shoreline that was kind of a decision based on game play we want rafts to be able to dock near the uh near the shore and and also be able to put those wraps to sleep in terms of physics updated and whatnot for the server so it's not constantly having to process those but um we did have a prototype it may actually end up in the game in the future of just kirsten or wave shoreline waves that looked pretty good and we had particles coming off then and whatnot so but the gameplay aspect that is difficult just because of transitioning from on land to the water [Music] so yeah long story short it's just uh for the shorelines it's just a bunch of gerstner waves with phases and amplitudes and wavelengths calculated based on the spectral parameters it's pretty easy to calculate but we decided not to put it in the game

yet because the players will not be happy uh trying to undock a raft or something uh with the giant curse or short waves just yeah we're working on a solution for that but yeah it's not in there yet more questions one two three sold sweet thank you