Shoot for the Sky: The Ambitious HDR Time-Lapse Skies of Forza Horizon 3

[Music] thanks a lot for coming along today I was wondered if you had to request that sort of funky music before it starts but it just happens automatically yeah like a couple of reminders before we get going I'm sure by this point in the week you all know these things but to begin with could you please ensure the volumes turned down and your Nintendo switches and the other thing is like don't feel the need to record this with your mobile devices it's all going to be on the vault and you know it's being recorded and hopefully they're recording that 4k checkerboard upscaling rendering talk as well so we can all catch up on that later date okay let's get started my name is Jamie wood and the lead lighting artist at playground games and welcome to the ambitious HDR time-lapse skies of

Forza horizon 3 the fort's of franchise has been around since the original xbox days and is owned by turn 10 who make the incredible forts of motorsport its game series at playground in the UK we create the open worlds Forza horizon games so to date the studio has released fortunate horizon fortune arisin 2 & 4 2003 and it's the sky technology of four to horizon three that I'll be speaking about today the technique that we used with it involves a lot of practical photography as we shoot these high-resolution high dynamic range time-lapse image sequences which we used to project onto the sky and inform the lighting on the ground and I would say there's there's a lot of talks at GDC that will claim to speed up your workflow improve your pipeline save your company lots of time and money and this talk is definitely not one of those this is going to cause you a whole lot of

pain and additional work and yeah a whole bunch more effort but hopefully by the end of the talk you understand why we chose to use this technique and what it brought us in the game so to begin with I'm gonna play the trailer for anyone that's unfamiliar with the game and just to give an idea of what it's all about [Music] [Music] what do we get game [Music] me

three [Music] [Applause] [Music] [Music] [Music] [Music]

okay so the game was released in September of last year and everything inside this presentation the screenshots and videos are from the Xbox one version of the game we also released on Windows PC and what I'll go through today then is first of all why we chose to invest in the sky why we felt that was an important part of our game I'll speak about the practical photography process of shooting these skies and how we clean up and process that data but if you're not into photography don't worry because I'll get onto the game side where I will explain the animation of the sky and all of the important lighting benefits we get from leveraging that sky data and finally I'll speak about our adventures in Australia capturing the Australian skies for our game that was set in Australia so why is the sky important to us well there's a number of good reasons first of all the sheer screen space that the sky takes up in our particular game

so we're a driving game we often talk internally about the car of the road and the sky being what makes up the frame and they're the things that we need to really nail and the you know the sky can be 1/3 or 2/3 of the screen space we also in the horizon games have built up a reputation for having dynamic time of day and whether that's something the fans really enjoy and that's something that we always want to push and the sky's part of that the sky also informs a lot of the lighting so we drive a lot of parameters like from that base sky so getting that realistic base is really important to us and finally like the sky really supports gameplay in a number of conditions so we have the action-packed kind of career that we have a bit more control over like dictating what times of day and weather are used in particular races but we also have the chilled out free roam gameplay that these picturesque backdrops really support so I'd like to speak a little

bit about our previous system unfortunate horizon 2 we had a a system for a represent in the sky that you'll see common and still a lot of open-world games today or anything with a diamond dynamic time of day cycle I should say and that was a base sky simulation that is based on the sun's position and modeling the scattering of light to create the kind of gradients you'd expect as a base at different times of day and then on top of that we would layer cloud sets that were particular moments of clouds particular shapes that were created by concept artists in our case and that would be like the cutout shape of the cloud the way it would fade in different layers to dictate how we would pump in light values to change these clouds over time of day and in our particular implementation we had 128 animated curves to control per lighting for our

cloud set so that was a lot of onerous information to handle they also you know that there's room for error there where we're dictating things like atmospheric fog ambient light all kind of artist-driven sets of curves and what that gave us then like if you had a random chance dice roll to have a cloudy day you might see the same clouds throughout the whole day and there's not really any dynamic change in the shape of the clouds there and you all you get as the the different artists driven lighting that's input to that so but we achieved some really nice stuff with the system so the game was critically acclaimed everyone really liked the visuals of it and and we we got what we wanted out of it but we decided for for 2003 that we didn't want a nice imitation of the real thing we wanted the actual real sky we wanted all the

layered complexity that you can see in these images that's just really hard to model in that previous technique we also wanted these massive volumous clouds with sharp lighting and cloud fronts we wanted killer sunsets we wanted these one-off natural occurrences and yeah we just wanted this this complexity from nature in the lighting and layering that that we hadn't seen done in any simulation or mock-up so we started to talk internally about why not just play a video of the sky which sounds like a nice simple idea but when you start having conversations about that there's some very good objections for instance the disk space like one hour of 4k video would fill up the entire blu-ray we need to get the rest of the game on there and one hour is not really enough we also have this big open world that we drive through it 200 miles an hour so we're already putting a lot

of strain on streaming so having like a heavy video asset in the sky isn't really gonna make people happy and we also didn't want to take a step backwards an image quality so we're used to these textured assets that we can create that a shark we don't want to add video compression into the mix and when we started to talk about a still frame sequence it was much the same trouble we assumed he'd need too many frames to be able to do this to have a smooth animation and so that that seemed like it was impossible to and there's a really good reason as well in terms of production where we've got this sky system that worked we've just shipped the game that everyone loved why would we add risk to the project and and try and change things up but we were really set on this real sky we wanted the real thing so we set about going out there despite our the warnings from the rest of the team and we thought right let's go and capture some data and we'll work out how to make it work so that meant going

outside in the in the winter and the UK experience and things like fresh air and natural light that game developers don't usually get a chance to experience and this process was really starting out with testing individual cameras just for the suitability of shooting these long time lapses and then it developed to that shot where I'm standing looking quite miserable in the but it's because I'm really miserable at the time and that's kind of the more developed Rick and even at one point they gave us some accommodation in the form of that motorhome on the end no Lamborghinis there so the actual photography side of things that has a lot of challenges with what we were what we were attempting to do so if you're a time-lapse photographer or if you're into that at all you will have heard the term a Holy Grail shoot which is what

you they refer to as like any time-lapse that goes through either sunrise or sunset because of all those changing light conditions it's called the Holy Grail because it's really difficult to do you've got to get good exposure settings and you've also got to kind of post process that to give you a smooth transition we wanted to do two of these shoots in HDR everyday so that we'd have a consistent kind of game asset we didn't want light loads of different like random exposures that will you know come back to us and a bit of a mess and every sky beat sort of different and graded differently and so that's that's a big challenge you also have this contradictory thing between like trying to expose the clouds quickly like and have nice sharp clouds but also we wanted to capture the full dynamic range of the Sun and sky so that meant we were using neutral density filters to reduce dramatically the amount of light coming into the camera which directly

contradicts trying to take a quick shot that's well exposed so you have to balance these two things together and also things like lens flare so having a very tiny aperture will actually increase the amount of baked lens flare that you're getting in these shots so you have to find a sweet spot there we were also using a multiple camera rig because we wanted a big future-proof resolution so that meant there was more things that could go wrong so the the actual rig that we ended up with as I say it's a three camera rig and we have this portable battery it's just a recreational vehicle actually at the bottom of the rig that supplies mains power to these three cameras I mean we used Canon 1dx cameras like I don't want this to be like a shopping list for going out and doing this like there's many ways that you could do something similar but we used the one DX with some 15 mil fisheye

lenses to get our fuel coverage and the reason one of the reasons we picked the one DX was because it has such good internal bracketing so it shoots at 12 frames per second which means you've got very small gaps between the the exposures that you need because we take seven exposures on each camera every 30 seconds so that was our our process and another advantage of the one DX is that it has two memory card slots which means we could shoot entire 24-hour periods without interfering really and you really need to get used to these dramatic changing light levels and how to handle that so it's an absolutely amazing education for any lighting artist to just get out there with a camera kind of experience what this planet does as it goes from day to night and I definitely recommend it so the cameras were monitored at all times we

didn't find a way to leave this rig in a field and come back three weeks later and pick up all the skies this was yeah a very active shoot and the way we dealt with all the changing light levels was to have a fixed aperture of f/8 which was kind of a sweet spot between that lens flare and sharpness and then we'd just ramped ISO to deal with the changing light levels so someone's there with the light meter reading how much the lights changing and then adjusting the cameras accordingly and one little trick that we did was because we needed these neutral density filters in the day and to stop down the light and get those exposures of the disk of the Sun we didn't want to fiddle around taking the filters out for a night and we didn't want to leave them in because then we'd have to use such high ISO to to the night sky so we actually swapped the lenses physically which was faster than fiddling around at lenses so we had

two sets of lenses and I put this image in just to show how kind of like low-tech this was like I'm really proud of how clobbered together this looks at this stage so we had our art manager turned electrician for this there's really just like a small group of us trying to make this work and so he bored all the holes in here and set up this transformer to supply mains power to all the gear and our Head of Production as well as like really into astronomy so he could recommend like jus heaters to reduce the condensation on the lenses and yeah whilst we've got a picture of some exposed eye electronics I should mention that we we didn't shoot in the rain so we decided that rather than tries to come up with something like because people were already laughing at us in the studio so we didn't want to make like a regular had like hair dryers on it trying to like deflect the rain or

anything so we decided that we'd shoot skies that looked a bit rainy and then layer that stuff on digitally later on and I have to mention the the production effort here because when we started thinking about these kind of big time-lapse shoots and trying to look for a location we really realized that we needed like a producer on this we spent about three months even finding like a test location that had like 360 degrees and all directions no light pollution which is really difficult in the UK where we're based because since the Industrial Revolution everyone's got streetlights everywhere and you know just finding things like secure land that you have permissions to use was something that you know really was a lot of work and not something we anticipated so yeah just to summarise the actual practical photography side of

things as I say a big production task iterative process to get the rig right it took us about six months from the first times we were out with a camera to actually having our production ready rig and it generates a huge amount of data to process which brings me on to the next section so we use two programs to clean up and process the data and get it ready to her game so that's Lightroom for cleanup and nuke which is a compositing package we use that because it's like a node-based system that we it's quite adaptable to whatever you need and and that kind of gets the images ready for the the game format that where we want to output and yeah each shoot with the seven exposures every 30 seconds generates 20,000 raw photos so that's one point two terabytes

of data to handle every time video shoot and this image here you can see like the dark to light exposures so we take shots like that every 30 seconds and that's about half an hour's worth of one cameras data so you can imagine what a daunting prospect is looking at all the images from three cameras from 24 hours and we knew at this stage that we had this challenge of getting this one point two terabytes down to less than two gig which was what we plucked out of the air as the sensible like you can probably fit this on the disk if you make it less than 2 gigabytes so we had a long way to go just to show you what one of the images looks like then so and you can see like the spherical distortion and down in the unused part of the image in the bottom left we we attached a diffuse light probe to the rig so we we always

have this point of reference of the the lighting and I'll speak a bit more about that as we go in terms of how we use that so in Lightroom we batch process the chromatic aberration and Vinnie from the image so that's just using a lens profile that you can download to suit the lens and we also do quite simple color grading of this so we all we do is we have 2x right color Corrections one for when we have the neutral density filter in and one for when we don't so all we're doing with the grading is kind of grading out that neutral filter that isn't exactly neutral and then we stick with daylight white balance on the cameras throughout the whole shoot and also we use Lightroom to clean things up like spot spot removal and dust our insects on the

lens all this organic stuff that we're not used to dealing with and sometimes even people's faces if you don't go out the way in time when it like shoots so looking miserable again so then we're ready to move on to nuke so we bring in our cleaned up images and we move we have these like sets of you know seven image sequences per camera and we merged them into an HDR source for each camera we unwrap that spherical distortion and Stitch the images together to give you something much more familiar as a lat/long map that you'd use on a sky and we also remove the stars here as well so we we found that no matter what resolution we use these in the final game you don't want pixel stars because we have my photo mode in the game so you can use like a really long lens if you want and take a zoomed in photo and you don't want this stars to turn into like

big squares behind the subject so so we decided to remove the stars from the images in nuke and then we have a star field that we plot as the replacement for that and we also create a cloud mask which is really based to show when to show through that star simulation so we need to know where we're allowed to see that and where this cloud that should occlude that so yeah and we also use nuke in the light probe analysis which I will get to so one other little trick that we do which is probably quite particular to our racing game is that we bias the resolution to the horizon so we're driving along most of the time we're interested in that bottom half of the sky and so here you can see the the top image is traditional kind of lat/long

format where as you go up the vertical direction your you've got an even number of pixels per degree as you go up we in our single down sample as we get the frame ready for game we just bias that resolution to the horizon so that we've got almost double the vertical resolution and there and one other trick that we do which I don't have a slide of but we we decided that because of that importance in our particular case where the the bottom half of this guy is much more important we actually cut the image in half and quarter res the top half of the image so we save three eighths of the footprint of this as well so yeah the data processing stuff is you know a big deal we we had a big IT effort on this as well so we had 80 terabytes of mass storage replicated off-site we had artists workstations that were had to be

equipped with large capacity SSDs for those nuke networks where they're very read/write heavy and as much as we streamlined the cleanup process as much as we could there's still a lot of manual work there just because of the nature of all these images so we we had to hire our full-time artist for sky cleanup on the project but now we have this asset ready to test so we've got this thing ready to go in game so let's have a look at the comparison so opus was the code name for Forza horizon 3 so that's what that word means but here you see again the Forza horizon 2 process where the the cloud shapes are very static they're just moving gradually and then you've got all these artists driven values for what those clouds will look like what the ambient lighting on the ground looks like and

here's our first field a test of the new system and what that looks like in game so you can see the moment-to-moment variation the the way the lighting on the ground always matches perfectly with what's going on the sky we've removed that kind of margin for error and you can see there's there's a whole lot more value moments a moment in this asset so at this point we were starting to convince people that this was quite a good idea so we still had this challenge to solve where we we have this image sequence where we're going between these nice moments in the sky but we knew that we needed to keep that framerate really really low in order to have any chance of actually shipping it on a disk so the next section talks about the motion which is really the key part of doing that use a low framerate and translate

between the frames in a clever way sir and the first thing that we tried then was a perspective correct dominant single direction of movement on the sky so this debug view has longer lines overhead where this represents like a faster distance traveled between frames so that's like the pixel motion that we have input to drive one frame to the next but this this motion kind of disappears to a vanishing point and slows down as it goes further away from where the cameras position so this is generally correct like that's sort of how skies moved if you go outside and have a look but it does rely on artist-driven curve again so we have to in for a direction and the speed that we want this sky to move in and if you get that wrong any particular time of day

you'll get this kind of sort of zigzagging effect and that's not very natural and the other problem we have is that a sky has a load of contradictory motion built into it so for instance the the Sun it doesn't really care what the clouds are doing right it moves on its own trajectory it is you know it doesn't care about this wind direction even but the Stars rotate around the pole star the moon does whatever it wants on an evening and you know you can't just drive it all with this single direction so we started to develop a system where we would track the Sun which is what this red circle is so that that's like us keyframing where the Sun position is and then we would mask away that cloud movement so that the Sun and everything around it in a sort of feathered blend would move along with the Sun so what

that looks like then is sort of a bit weird because everything around the Sun is very static and you get this sort of stretching away from the Sun and yeah like sometimes that that's not really the correct movement there these clouds could have been moving quite quickly in a different direction to the Sun and having them pinned like that really didn't suit what we needed so we started looking at other methods and what we settled on was using motion vectors to drive the motions so what this is then is we pass in our full image sequence into nuke and using nukes built-in vector generator node we would output this set of textures and for anyone who's not familiar with texture driven motion this is a four channel floating-point 32-bit texture where two

of the channels describe UV animation from one frame of the sequence to meet the next frame halfway and then the other two channels with positive and negative pixel values will move the frame back in time to meet the previous frame so there they're describing actual texture space motion in each channel and this is an example so this sequence is a two minute gap per shot so that's a very jumpy thing that doesn't really look like a video that you would want to play in your game but the same number of frames just with a motion vector driving the motion between them gives you a much smoother interpolation it's kind of local enough that it actually describes some more of those kind of smaller clouds and how they're morphing and

changing for the next frame and it means that we can play this at any frame rate we want because we're making the in betweens up so we can play this at 60 frames if you want just with those one shot every two minutes and if I've done my maths correctly that's zero point zero zero eight frames per second was the frame rate that we have there so that's you know a significant saving from something like 30 or 60 frames per second with the number of images that we need and we also discovered through testing that we only needed a quarter res motion vector so we didn't see much in the way of diminished motion by quarter rezzing that so that was a saving there and we we made a further saving by instead of storing the fill 32-bit image we would keep an XML that went along with the frame data so each

frame would be given a maximum pixel movement which was found per channel in the original motion vector and we'd map that 0 to 255 on an 8-bit texture which is the or pretty-looking one below and and then when we're back in game we can look at the xml and we can remap those values to the original range of motion per channel per frame and that was like a really nice optimization because that texture compresses much much better in game so to summarize the emotion side of things then you know you've got to solve those contradictory motions in the sky the way we did that was with the automatic motion factors we also had the ability if anything ever kind of didn't pick up correctly we could easily input on that original frame data like markers that Nick could pick up easier so if for instance the

motion of a moon didn't quite pick up we could you know draw a big red blobs over it and then nuke will kind of track that a bit a bit like our original methods with the dominant direction stuff so the motion factors were the way we did that and this is really crucial to being able to use the whole technique without a low frame rate this doesn't happen and I'm sure this is that's probably something that's put other people off this kind of technique before so I'd like to speak about the most important part of this whole thing from my point of view which is the lighting benefits in the game so it's not just about what's going on on the pixels on on the dome this is we thought we saw some real benefits in the game lighting and rendering as well so to begin with in the light probe calibration they're promised so you can

see in the top of that image you might have to see the original gray ball there the light probe that we have in every photograph and so what I've done here is just cropped in on that image like just the pixels of that ball and because in the shoots if if we're in the northern hemisphere we face the the gray ball to the north and if we're in the southern hemisphere we face it to the south and that means you've always got generally normally you've always got a point of reference on this ball throughout the day from some science sunrise to sunset and so I've plotted the max luminance found on that sphere here in the graph below and there's some really interesting things that you can see from that so the sort of wavy break up on the the first part of the graph like so that's when the Sun rises that's because there was like sort of broken cloud in front of the Sun and it's just sort of diminishing and changing the

intensity and then it kind of levels out to a more consistent relationship between the RGB and then you've got these like huge dips in intensity which is when the the sun's actually behind like full cloud and you drop down to just the ambient levels and that's that's really interesting because that means like the rig itself was it in full shadow there and that means that our player should be in full shadow when they're on that part of the image sequence so we can use this data to model that in game and give a real kind of true idea of what the lighting was like at that time when we captured this guy so just to give you an example of that then that's the original photography and nuke on the left hand side and in pics which is one of the Xbox development tools that we use you can see in a test bed the full intensity of the Sun on the bottom and then where

we've knocked that out with cloud shadow in the top so we don't we don't ever change the intensity of the Sun to model this we do it much like real life where you actually just knock that out with the shadows so this gives you like a really nice balance between the Sun the sky and the lighting and thanks to the hard work of the environment team at playground you can see the balance between so the images on the left are our near final Australian track and the images on the right are our actual sky captures and you can see that that relationship really comes through and it's a realistic kind of balance and that happens throughout all times of day so the tricky kind of low light times a day where you're you really depend on that that relationship between the sky intensity and the ground to look really realistic that that all comes along with

this kind of more scientific approach of analyzing what the what the lighting is doing and because our time of day cycle is now an image sequence or several emetic sequences we can make other images to go along with those sky frames so one such image is the fog cube map that we have per frame so this is a simple Gaussian blur of the sky that is flipped to give some negative directions if you're looking down and you want to find a fog value and we actually analyze this image before creating the cube map find the brightest point which we assumed to be the Sun and we clone that out automatically in a tool and this kind of blurred 360-degree cube map gives you a really nice fog color as you disappear into the distance so we have

that's actually a screenshot from Fort Sill horizon 2 where I've used the 4003 system but you can kind of see those distant mountains just like matching up perfectly with the sky in the distance and that's like a replacement for directly an artist driven thing where we would have been picking a fog color and we would have had practically no chance of matching that all the time - whatever clouds or simulation simulated gradients were in the sky so that's really nice and on a similar note the we have this diffuse sky indirect cube map so we take the sky image with an infinite black ground plane and we convolve that offline so we've got the opportunity to do really nice high-resolution m cube involves a 180

degree cosine blur and that is just the indirect diffuse contribution of the sky as if kind of you're standing on top of a hill and there's no ground so it's like not really a practical thing to have that's never a scenario that we have but what we do is occlude that with our voxel based GI system so we have voxel based AO that will reduce this diffuse skylight and then we get the rest of our diffuse lighting from the voxel based radiance that system kind of backs this up to ensure that at all times of day we have this nice relationship in the shadows between the sky and the indirect lighting so as you can see in those images we've got this kind of automatic confidence that those levels match the sky another texture

that we have is the cloud mask as I mentioned earlier so that's used to reveal where the stars are seen or not seen but we also use that on the distant cloud shadows so we take this black and white mask we turn it into a cube map and then assuming that the clouds are all 2 kilometers overhead that's just a number that we used we can look at any distant terrain pixel you know project up to where the clouds would be and then look up as to whether that's in or out of shadow and that we found that to work really nicely so the the having like a relationship clearly between the clouds that are in the sky and what's on the ground felt really natural obviously in an ideal world you'd want to kind of model these things in 3d and raytrace through a shadow and that always wouldn't be that what you see in the sky

is projected on the ground but having that relationship seemed really natural to and because this moves along with the player like the sky has to follow the player as it moves that meant it wasn't really suitable to be used at the car so you don't want to as you drive all the clouds move with you because that's really weird so we have a scrolling texture solution for up close so this video is really sped up so that you can see it or maybe it's playing really slowly and slowed down but you can see the the clouds coming and going there around the car and that's a different system entirely so we have a scrolling texture such as the one that's there and that will move in the direction that we input that the clouds are moving in so that is a curve that goes along with each sky and that means that you can drive towards clouds they can come and

pass over you and all the things you're used to seeing when you drive in the real world you'll get with this and we can also vary the the cloud shadow at car intensity based on that original graph that we draw from nuke where we can make sure that when the rig was in shadow the players in shadow using this system and I wanted to mention the weather which we developed for forts of horizon 2 so the weather system was something that we worked really hard on for that game and it plugs into this new sky tech as well so based on the porosity of the materials in the world will we're able to tag with the sky sequences that that come along like this this point here has a chance of rain and then that will happen really naturally as the clouds come in the ground gets

wet the sort of weather systems move on and then you get those nice post rain moments as well so all of that stuff just kind of plugged in really nicely and one other benefit to have suddenly having a constant with these photographic skies so we've got this these images that we've controlled we've taken all the exposures we've got the metadata for how those skies were captured that means we've got the opportunity to take other photographic reference and seek that in the global hierarchy of lighting next to the sky to give a really nice relationship so one example of that is just seen in this this test here so this is just early pre-production stuff and that's some images I took from my car and turned into an HDR texture but we can see at this twilight periods that things like your your gauges are starting to come alive as you'd expect as things get dark

and then headlights are blowing out brake lights or seeming really hot and then as we move to a daytime condition where there's a lot more light the game cameras exposure has had to adapt to that at the same texture appears naturally to change as you'd expect so you can see that those gauges recede the headlights are less blown out in the day and the same with the brake lights as well so that was like another advantage of suddenly having this kind of known quantity of the sky so yet to summarize the kind of game side stuff it's certainly as big a part of this whole technique as sort of what's going on in this guy on the sky dome we benefited an awful lot on the kind of jump in realism just from using the sky technique based on what what the other stuff that was

going on on the ground so finally I just like to speak a little bit about our adventures to Australia where we sort of decided that whilst we have this rake and process that can capture real skies and our game is set in Australia then we're better to shoot the skies so we we ship the whole rig out to Australia we went on location me at me in the art manager and we trained up to photographers out there who did a really great job over the summer of 2015 so that's like December 2015 the year before we shipped and and over the course of the whole summer they they captured 25 full 24 hour days and that was like our pool of data to use as our skies for the game they took two million photographs it's just a big number

not really significant just sure enough so this is a video of us out there and it was really like something I'm really glad that that we did like go out to the real location it's like a really romantic idea to kind of take the lighting from the actual real world location and kind of transport that to the player wherever they may be and as well as kind of just that being like a cool thing to do we got the the Sun arc included in that data as well so normally we would kind of go to great lengths to kind of work out the timings of the sky and where the Sun would be positioned and how long sunsets at last and like we got all of that stuff in the data which was like just that other level of realism and we also have this like survey of all the weather in Australia so we'd been there for the whole summer we knew how long it rained

for and when it was cloudy and when we had these full clear days and we could use that data when it comes to setting chances of these things in our weather system in the game so here's some of those Australian skies in our near final Australian track then so these are obviously sped up to look cool as time lapses but this stuff would roll at a natural speed in-game and gives you that kind of nice feeling of things just evolving without you really noticing like much like he do and you kind of spend the day outside and then all of those images that I showed you at the start where I said we wanted to capture the real sky they were course things that we captured in Australia all of those things made it into the final game and that gives us this kind of wealth of options when it comes to setting up the the races or online events or whatever it may be we've got this like really nice full

range of different moody dramatic conditions some shots from the DLC there and I think that that final shots actually from one of our players from our community competition which I really liked and yeah just to kind of sum the whole thing up then having been through it it you may have gathered it kind of represents a huge cross-discipline amount of amount of additional work to do this like it we definitely touched on all departments I think from art engineering design with the way they set up the races you know the leadership team like taking on this kind of crazy idea the IT effort the production effort and you know not least the my partner in crime on the rendering sites the rendering engineer that was assigned to

this part of the project did like a crazy good job of implementing all of this stuff and really wouldn't have happened without without him so there's also this quite huge investment of effort on the practical capture side as well so it's almost like this new department we've got where it's almost like an audio team or like getting into photogrammetry or whatever like you've got to kind of understand that that's kind of a huge investment there but the sky' system as I say has really benefited the visuals and a kind of jump in realism and it's removed all of those artists driven data curves that I spoke about earlier on as well so we had barely any meetings if any during the project where we were sitting scratching our heads and going are there's something a bit weird at the lighting what is it like because it was done in a sort measured kind of scientific manner so we

had sort of confidence in it as we went along and finally you know whilst obviously shooting natural skies and doing like time-lapse photography shoots that can be rained off as you know adding a whole lot of unpredictability and kind of risk on the production side of things that unpredictability is also kind of like the biggest benefit of this like nature's in control you've got this kind of wonderful performance of the clouds and the lighting and that's really the strength as well as the system like having that stuff captured and preserved and being able to show that in the game is something that I'm really pleased that we managed to do so yeah that's everything thanks very much I think we have time for questions doing

no one seems to know oh I'll check yeah I think we have like ten minutes so if anyone has a question then great Todd think thank you did you consider shooting HDR since using it for lighting so the images that we created were HDR sources yeah so like the the exposures were later combined to give us the full dynamic range actually of the southern sky I mean we we actually didn't use that full range in the end so that's something that maybe we did differently like the it's not always a benefit to have that baked through the intent Sun in the in the image so you're sometimes better like offloading that to the key light and but yeah this is the images were where HDR yeah so how did you use that HDR would you have normalized it and then used the camera six the game cameras exposure to so I think just

simply not having such a hot extreme Sun in the image would be actually slightly improved so not I wouldn't like we were very conscious not to kind of tinker with the data coming in we wanted a very kind of linear kind of process from the raw photography right up to getting into the game so we didn't want to do anything like grading on top of it or anything so I think just you know be different for every game but just being very conscious about how how much range you actually want in there because it depends on kind of your rendering pipeline as well whether you're actually clipping out like offering this like huge intense hot spot on the image right so you didn't wait balance it or anything you just took it off just yeah just daylight balanced shots and then white balance as a kind of process when

we go through lightroom just assumed it was shot as daylight and then you know you've got the option in-game then to white balance however you would like for yeah cool thank you thank you sure a good doc I got two questions one of them is like how do you define the final value of the direction lights in the game and the second one would be how do you pick the final exposures for the in-game especially at night time where you know designers sometimes well the real life is too dark yeah good question yeah so the first one was the Sun intensity so that's really taken from that that light probe data so where I'm analyzing the max luminance found on that sphere that's that's not directly what you would want as your son intensity because that obviously includes like the ambient light but we would model that not really by using the

exact value we found in that graph but by using a modeled like PBR diffuse light probe in our game as well so like I'm able to so we had some debug options so that I can mouse over the screen and see like the exact kind of linear and final values on the gray bar and I can compare that in the original photography to the the values I'm seeing there so it's kind of like a diffuse plus key light equals this and then I'm able to balance it that way and what's the other question the final exposures like for night time because like horizons and I compared to like daytime is it completely different yeah value so we actually kept the original difference in the images so that was kind of hard to fit into the texture format so the

render and engineer on this project actually had to do a kind of scaling factor to fit it into the texture format because of that range being so low in like your night shots are very very small values when when you save it so we actually boost them up to save out the texture and then we use them back to get the full range back so we didn't actually cheat night we did have a very very dark night like if you if you play the game it's it's really like headlights only that you can drive with which was something yeah the art director really wanted that and so we tried to stick to it as much as possible we did we do have like some ambient controls so we can use the underlying sky underlying star simulation to actually contribute to the diffuse light but I kept that really to a minimum like we've pretty much kept the really dark nights actually yeah thank you thank you yes hi fantastic

talk thank you for sharing that and my question is you short time lapses at 30 seconds intervals but interpret using two minute intervals yeah yeah that's right so I think shooting at 30 seconds is a really good idea because wind speed actually really affects this you can imagine like the amount that the kind of speed of clouds varies it's like sometimes it'll be easily like four times faster and therefore you get like in terms of what you'd ask in the motion vector to do you would be wanting to move you know four times as much width so the original kind of frame gaps doesn't always relate to the amount you're trying to move the cloud so our target was two minutes that's what we found to give us the zip size we wanted but the worst skies that had faster wind speed that we had to use it one minute intervals which kind of 30 seconds gave

us that option to do that and I think there were some skies that we managed to play back yet much bigger interval so if you have a clear sky without many clouds and all you're really doing is kind of tracking the Sun we found you could use like much bigger and you know frame gaps for that so we used it variably but it's really done to win speed more than kind of time lapse intervals that dictates that yeah because it could save you a lot of space when you when you know the end of interval that requires certain sky and also my second question is can you say about technical details how much what is the end up resolution for full time lapse and how much it requires and a disk space yes so we we used 4k final textures in the end that's obviously

with the three three eight saving where we down raise the the top portion so it was like a 4k by 512 and then a 2 K by 256 texture so that was our final resolution that we used and on disk we ended up using a about seven gig of the disk budget so it's pretty significant commitment yeah yeah and my last question is you you have a dynamic weather system it means that you can mix difference guys like friends completing the filter time lapses yeah so we we had that system so we developed a way to put blend points and we could have as many blend points as we like between the different 24-hour tracks we actually kind of reined in how much we used that

because we we found that we really loved the kind of natural transitions that we were getting and when we when we used our kind of manufactured ones it felt like we were kind of putting part of it back to that previous system where it's like sort of made-up stuff not real stuff so we we tend to only use that when we needed to for instance like obviously at the end of the day you have to kind of blend to a different portion so yeah we had that system in place we didn't use it as much as we thought we would thanks thanks okay that's great what how many skies did you shoot and what was the selection process like did you have reviews on the time-lapse and yes it was shot 25 Australian skies for the pool of skies that we used in Australia and we shipped with for like we reduced that down to four 24-hour

periods so but that's you know you can see from the video where I showed the comparison between horizon 2 & 3 like there's so much variety in each day that 4 was actually quite a huge number like we had a whole lot of stuff to choose from in that and the review process was quite difficult because you if you see the format that they come in when you've got the kind of spherical distortion and the single-camera sequences and they're all in bracketed orders as well like it's it's very difficult to get an idea of what that would be like to play so we we develop some tools to get around that so we would in it in every raw image you can actually extract a JPEG which is embedded in that raw so we automatically found the brightest exposures which are normally like the well exposed ones and we'd extract the JPEG sequence and then we'd automatically kind of write that to a video but it was still quite

challenging to kind of know that you were making the right choices because you're you're still kind of flicking through these big big expansive skies when you're going to be looking at this kind of little part of it on the horizon so it's definitely something I would probably improve and if we do it again so my second question is are you going to do this again I thought I got really speak about anything in the future that the studio's really do sorry thanks yeah I think the last question then cuz we're great I'm lucky it's the last great talk the best of the week for me so it was I learned a lot of things and I do HDRI myself and we get some kind of problems when we are making HT re in HDR eyes of the skies because as you mentioned you have to 12 all these time-lapse you have to make some motion

vectors to make in between motion yeah to get what is missing but also in the bracket things in the different photograph you have seven the time is not exactly the same yeah and you depend a lot of the shutter speed and because you are using audacity filters sometimes the dish shooting time is quite long do you have you do you have some advice about this yes we didn't do any like offsetting of frames as sort of clouds move like which I've seen people do where they sort of try to adapt for the emotion on each exposure I mean our exposure is because we're using there the 1dx of exposures were less than three seconds even though you know that's still quite a little okay a bit of blur though but I think we we kind of survived because our source resolution was so huge so like with the three

cameras we end up with like a 13,000 pixel source so you know by the time you kind of crush that down to 4k it's those those small movements or you know so it's because it's just three second that this doesn't matter so much yeah I guess I mean are you using longer times than that when you do your HDR eyes yeah yeah so probably we we manage with that yeah yeah do you change the setting you considering the the lighting condition the cloud and speed the wind speed there is a lot of wind maybe you try to shut faster so we didn't I mean we you know that would be quite nice in one way but also we wanted to keep it quite a you know structured processed thing because we were going to shoot so many we wanted to make sure we

always had like the the same exposures coming back and like no surprises when you're cleaning it up where someone's like maybe quickly taking the nd filters out and then done quicker shots because it's all overcast or something and so we opted for that consistency over doing that but yeah you could definitely get across so the advice is to find the right setting for quite all the situation and try to stick to this for the better for the best workflow and I think that's probably quite specific to our needs on on the game so like yeah we you know we wanted these consistent things I suppose if you're if you're taking a time-lapse of fast-moving clouds and you know it's fast-moving and you have time to adapt to it I think you'll probably get a better quality right from doing that but yeah okay thanks alright

you