Interview with Juice on 'ARES: Our Greatest Adventure'
When you are left behind on the second smallest planet in the solar system it usually means one thing. Unless, you are Matt Damon starring in the science fiction blockbuster The Martian, produced in collaboration with NASA themselves. For production house 3AM, Polish post-production studio Juice made a three minutes long teaser which explains the scientific basis of the ARES mission and events portrayed in the movie. We wanted to know more and interviewed Juice's rendering team.
Can you tell us a bit about yourself and Juice?
Jakub Knapik: Juice is an awesome company driven by design oriented artists, with a great home-like feel in the offices, divided into two main locations: Wroclaw, the bigger one, and Warsaw, smaller. It was the second one that was the main host for the ARES project. Juice also has a smaller office in Tokyo, dedicated to Asian markets. Juice was founded in 2006 by Wojtek Piotrowski, Pawel Janczarek and Adam Tunikowski. All three were designers coming from different fields of interest. Wojtek’s focus was branding and typography, Pawel’s focus was on print and illustration, Adam's was retouch and animation. Juice specialises in design, sound and postproduction.
I am a VFX / CG Supervisor and for the last few months I've been working with the CG crew in Juice's Warsaw office. In my previous jobs at other companies I was focused on VFX for feature films and special projects, and that is also why Michal Misinski invited me to Juice when the ARES project was brought in. I supervise but also love to work as a compositing, shading and lighting artist.
Artur Szymczak: I am 31 years old. I have worked in the industry for 8 years now, and I was a freelancer for the last couple of years. Since last year I work at Juice as lead shading and lighting Artist, but I have been using Arnold since I saw Daniel Martinez Lara's "Pepe" animations, so basically I am using it from the very beginning. Juice is the fastest growing VFX studio in Poland, keeping the great atmosphere of a very small boutique.
Seweryn Czarnecki: I work with the CG team in Wroclaw, and for the last 10 years I focus on CG for VFX. My speciality is shading, lighting and compositing.
What’s the production story behind this teaser?
JK: It all started because of the relationship that Michal Misinski, Art Director at Juice, and Ash Thorp, Art Director and ARES Director built up during the previous project Ghost in the Shell. Ash contacted Michal and asked if he would be interested in doing a few shots for his latest project for Ridley Scott's upcoming movie, The Martian. Obviously, he was far more than just interested and it all started rolling. During the work, we showed the first render/comp test of the only Hermes shot we had at that time and that opened up the bag of shots for us because of a very positive reaction from the client. It was a great moment for us and a huge boost for the team. We had a chance to work along the best of the best in the industry in a very special project with a rather limited technical infrastructure at that time.
What was the size of the team that worked on this project?
JK: The whole ARES team was rather small and had around ten artists. I worked with most of them before and knew the artists really well, while with others I really wanted to work for some time, so the whole mood of the project was simply fantastic. It is a brilliant crew. The Arnold team consisted of five artists: Artur Szymczak worked with me in Warsaw on Hermes, HAB and one of the Mars shots using Softimage and SItoA, while Seweryn Czarnecki, Andrzej Przydatek and Tomek Przydatek worked on the DNA shots and Mars descent shot in Wroclaw using Maya and MtoA.
Why did you use Arnold for this project?
JK: I consider myself a render geek, I have used many rendering engines, and obviously they all are brilliant rendering tools with amazing features, but for me while working on a project, it was always about getting on screen what I had in my head in the most linear way possible, so that I don't get stuck in the technicalities that would suck me in and chill my idea too early. That is why when I first found Arnold I immediately implemented it in a previous company I worked at and it was a great success. At that time, we were one of the very few commercial Arnold users around. I fell in love with the raw power and straightforward nature of Arnold. The fact that it uses a brute force approach, saves a lot of time in the tweak phase, because you just need a few quick tests to tell the right sampling levels, and it's all very predictable. I cannot say how much I love it as an artist using it and as a VFX Sup that has to keep the deadline and plan the work. Also the fact that we managed to have brilliant artists onboard for the project, whom also were used to Arnold and are die hard fans of it, made it an obvious decision to use it. But there was also one huge reason why we used Arnold. Hermes, a fantastically made 3D model created by Framestore, was, after intensive optimization, still a huge 70+ GB asset. We had to render it out at 2.5k and in some cases at 4k, having just workstations with 32GB of RAM, with proper motion blur, multiple GI bounces and extremely intense cosmic sunlight, that always brings some sampling challenges with secondaries. Arnold was the best solution for us.
AS: I am using Arnold since it appeared on the XSI platform only and I am a huge fan of the feeling it provides. It was basically "love at first sight". Arnold has a tremendously powerful ASS stand-in's workflow, which was really the key in ARES 3. It's very straightforward. It's blazingly fast. It handles very complex assets and geometries, it's very artist friendly. And of course I knew that Jakub had been using Arnold and XSI from a long time in his projects, so it felt even more safe to use it. And to be honest we didn't even try to render it out in different engine ;) We didn't even think about it.
Was it difficult to transition to Arnold at Juice?
AS: That's a great question, because ARES was the first project on which I had the opportunity to work together with Jakub Knapik. For 8 years I am using XSI as my main software and Arnold as main renderer. For me personally there was no transition. In these days it's really hard to find someone working in the exact same software environment as me. So for me personally working with Jakub was a bless. I was the only person responsible for the lighting, shading and rendering, so we didn't have to retransform the whole Juice team, we just needed to build the pipeline around it.
JK: No, not at all. The great thing about Arnold is that it has all technical controls simplified into very straightforward fields. You could say at first sight it is suspiciously easy to operate, but to me it's the intelligent design behind that just makes your life easier. On the visual side, it is a pure joy to use, and since it is so straightforward brute-force global illumination renderer with powerful shaders, you can just focus on shading and lighting. I would call it one of the most logical and easy to learn production renderers there are.
What were the main rendering challenges?
AS: The biggest rendering challenge was just simply to render it out ;) We had the model and textures from Framestore, so you could think that my task was to just setup the light and hit the render button, since we all know that Framestore's asset is simply perfect. But it wasn't exactly that simple. The main challenge was to render out the 700GB asset on our 32GB RAM machines. As soon as we saw the Hermes station with Jakub, we wanted to remove all the little screws, and the smallest elements, that you couldn't spot in any of the shots because of the distance from camera. But we didn't remove anything. We wanted to keep as much as we could from the 700GB monster :) So we took all the textures, downscaled them, we made TX's from them. For some of the Hermes parts we resigned from subdividing in the 3dsmax model and it still looked stunning. The geometry in the assembled scene was eating 10GB of RAM, and 21GB for the textures. When we started rendering, we hit 31.2 GB so we still avoided swapping to disk. It was the toughest rendering challenge in my life honestly.
SC: From the very beginning we had the hard task of finding the right picture that would blend the classical DNA look everyone is used to, and an organic look we all wanted to have. Also, we had to find an interesting way of showing dying DNA and its final destruction. Using alShaders with its SSS options helped us a lot in the development of the final look, and use of 3D depth of field to get all the nice subtle defocus on the particles created the final shot. I enjoyed every minute working on this project.
JK: I think every task had its own challenge. Hermes was really heavy from the poly-count and huge textures-count point of view. Also, we all had in mind that we will be facing a challenge of having our shots side by side with a stellar job done by Framestore that we did not had a chance to see till the end of the project. That meant we could not cut corners too much to fill our technical boundaries. I gathered a lot of NASA photos form space missions to match the lighting and shading behavior and we looked at it constantly making final tweaks to Hermes. We learned a lot about space lighting that way. The DNA shot was more of a visual search for the right look that would be organic and natural but first we had to find what that meant. It was more of a creative process with a lot of trial and error.
Can you talk in detail about the Hermes asset?
JK: We received the Hermes asset from Framestore through 3AM Company as a set of geometry blocks and a huge set of textures. I can tell that out of the box we were blown away by the amount of textures, I think it was far over ten thousand 4k UDIM textures and heavy gigabytes of pure geometry beauty and madness divided into big blocks in FBX files. We knew right away that we will have to work this out somehow to make any use of it in our environment.
AS: We ended up with 291 million polygons without subdivision, and 356 million polys with some objects subdivided already in render. We were subdividing only the biggest objects in the ship because of our RAM restrictions. We had around 2800 texture maps from Framestore, which we downscaled and saved as 16-bit files, so their weight dropped down from 339GB to 40GB.
JK: It was impossible to render it in our technical infrastructure at Juice at that time. We could not open the model at once obviously. Frankly we could not even open a third of the model into one scene as pure geometry, not to mention rendering it with any textures. We were simply running out of RAM. We analyzed the use of 2800 4k textures that we planned to use and downscaled some of them into smaller ones to reduce the load. From the geometry point of view, we went with finding the repetitive elements and after setting up the right shaders, Artur was converting them to ASS stand-ins. After that we rebuilt the whole Hermes out of *.ass stand-ins files. It all worked out, and I think when we finished we were able to render it on 32 GB and have a few percent of RAM left just to prevent the OS from swapping. For us it was insane and thrilling at the same time to work with such asset.
How did you approach rendering the Martian landscapes?
JK: We worked on two Mars shots, one from orbit when we see a small Hermes orbiting the planet and one when we descend to Mars to show the possible colony in the future. All other Mars shots come from MPC or Framestore. Funny thing is that both of our shots were approached differently from the pipeline point of view. The one with Hermes orbiting the Mars, was created using Softimage. We used a Mars heightfield gathered during one of the Martian missions, used color maps with aiStandard shader, added a layer of volumetric scattering atmosphere on top of that and rendered that with SItoA. Arnold worked with complex displacement maps really well. The descent shot with colony was far more complicated as we knew we will have to work harder on details. It was a hand sculpted surface that was rendered from Maya using MtoA. It was a rather heavy asset.
SZ: Our biggest challenge was to render out a really dense mesh in the Mars descent shot. We decided to go with a brute force geometry approach and skipped using displacement. We used 3D-Coat to sculpt the surface of Mars having photographs as a reference and we ended up having a sphere of over 100 million polygons. We were not sure how Arnold would deal with such high-poly count in a single object but it was just fine. For this shot we did not use any custom shaders, just a plain aiStandard shader and a set of textures to give the Mars surface the right colors and details.
Did you have to write any custom shaders?
AS: We didn't write any custom shaders, we used alShaders for Arnold, which are really great to work with. alSurface worked great in the project simply because of its 2 speculars. Hermes itself got some elements that have the standard Arnold shader applied. So even if we wouldn't use alShaders, we would probably mix some standards and still get the decent results.
How big was your render farm and what were your render times?
AS: Rendering times varied along the different shots. Most of the frames took around 2-3 hours. The longest one was the Hermes station flyby. We were hitting 5 hours per frame in 2.5K resolution at some point. We rendered some of the shots on Juice's render farm, with only 10 Intel 8-core machines, but the heaviest renders were sent to an online renderfarm.