Thu 26th Dec 2013, by Nadezhda Markalova, Polycloud.ru | Production
Stalingrad was released in October 2013 and will be released. After two years of planning, an audience could finally appreciate what amounted to years of work for the Director Fedor Bondarchuk, the DOP Maxim Osadchy, the Cinematographer's team and the actors plus the incredible work of Production designer Sergei Ivanov. And, of course, the monumental efforts of the VFX Artists.
Visual Effects for the film were undertaken by the studio Main Road|Post under the supervision of Arman Yahin. There are a total of 230 shots with visual effects in Stalingrad, many of them quite long, taking approximately 30 minutes of overall screen time. Various software packages were used - Maya for modeling; Houdini for animation, shading, scene assembly and rendering; MARI for texturing and Nuke for compositing.
Main Road|Post started work on the project in 2009 with the first version of the film’s script. Visual development began with concept artists creating mood sketches, to set the atmosphere and stylistic look of many of the scenes. Two years later, in 2011, the project went into production. At this stage many tests were carried out, including tests in stereo. It quickly became apparent that a number of key scenes of the film would rely heavily on computer graphics.
Several artists travelled to Volgograd (the modern name of Stalingrad) before any work was started on visual effects. The Battle of Stalingrad Museum was a key destination. Here, we shot reference footage of a miniature of the city. They were particularly interested in the site where the house of Pavlov stood as one important task was to reconstruct the architectural style, the peculiarities of the building and the character of the area. Many of the master shots of the city were scripted to be seen as views from the windows of Pavlov's house. It is worth mentioning however, that the area where the main action of the film takes place was originally conceived as a compilation of many symbolic places in the Battle of Stalingrad.
The opening scene of the film takes place in modern Japan after Fukushima. Several fully computer-generated shots were made for it. The concepts for full-CG shots with the arrival of the EMERCOM plane were developed first. The opening shot of the plane flying towards the audience was enormously time consuming as according to Director's vision, the plane first appears on a background of a huge red sun resembling the flag of Japan with the bottom of the shot framed by cumulus clouds. As the film was made in stereo it was impossible to use plates for the clouds so we had to make the clouds in-house, so they didn't look like flat plates in stereo. The work on this shot began from the moment the project started, but every time we went back to it, there was something that did not completely satisfy the team. The clouds didn’t look real enough. Unexpectedly, director Fedor Bondarchuk decided to reject using clouds altogether. He introduced a new concept with another color scheme, and when the clouds disappeared from the shot, everything fell into place.
The scene in which the Red Army soldiers cross the river in pre-dawn to the burning inferno of Stalingrad demanded very careful work with colour. The shooting of these scenes was carried out in the daytime, relying on subsequent repainting "by night" in post-production.
The scenes which included footage of burning people were very labor-intensive with work on them taking about one and a half years. During the crossing, the Germans blow up an oil storage facility and fire covers everything, the hill, water and people. The scene is long and consists almost entirely of shots which required extensive post-production work. CG was used to enhance key elements of the whole scene including, color, contrast, atmosphere and the appearance of fire and smoke. To keep stylistic integrity, detailed concepts were firstly created for all the shots. Then an animatic of the entire scene was constructed, allowing us to understand the rhythm and dynamics of the action.
Almost all the shots in the scene had to have fire added either in the foreground, the background, or everywhere at once, so it was almost impossible to use shooting plates in a stereo project. Large amounts of detailed CG flames had to be simulated. CG fire on the project was a very complex R&D task, with three months spent on the research, creation, and debugging. We did it by using physically-correct technology based on volumetrics. At the same time we devised a method to increase the detail of the finished simulation by applying displacement that raised the reality of the appearance of fire and smoke, especially in the foreground. Displacement greatly increased the rendering time of the simulation. Initially each frame rendered at 8-12 minutes, but with the displacement the rendering time was increased to 2-3 hours. Also the final result directly depended on the quality of the simulation. If it was bad, then the displacement wasn't going to save it. Displacement can only improve a good simulation! As all the fire and smoke on the project was created using this technology after receiving the first fire simulations, Andrey Maximov, Art Director of the Main Road|Post, composed several key shots to demonstrate the color scheme of the entire scene. This greatly simplified the rest of the work as it formed a unified vision of what the explosions, fire, and smoke needed to look like.
Fire was only one aspect of the work. For a couple of shots in this scene we had to make digital crowd simulations as it was impossible to shoot stunt men near the exploding tanks. All the crowd scenes on the wide shots are CG. We tried several things to solve this problem.
At first, we wrote a complex dynamic system in Houdini, which considered each soldier as an intelligent 'agent' with a set of behaviours, who would choose an action based on the situation in which he appeared. It was a simplified version of the Massive system, and turned out to be very effective. In all the shots where there was a lot of movement the animatics were made with the help of this technology. When the creative element was in use and it was necessary to change the behaviour of the agents for artistic reasons, they were animated manually. The system with 'agents' was very useful when there were 30-40 soldiers in the frame, and could even be used manually with up to 20 in a scene.
Another aspect of the crowd scene creation was the preparation of animation clips. A less realistic model will not be readily noticed at a small scale in a large scene, but unnatural animation will be noticed immediately! The creation and preparation of the clips with realistic animation was a complex and laborious part of the work.
The creation of the clips for the scene with the exploding oil tanks was further complicated by the fact that the soldiers were running over uneven ground. Usually, when a character moves on a flat surface the work can be reduced to a small set of alternating cycles of walking or running - here it was necessary for the model to adapt to uneven terrain. If a soldier falls into a ditch he must collapse, and if there is a hill is in front of him, he must lift his leg higher so it was no matter how many clips we created we were never going to have enough.
It was going to be very labour-intensive to abandon clips all together and individually animate each soldier, so special tools were developed using Houdini that allowed the animations to be corrected and adjusted based on the terrain changes. For example, if the animation clip had a foot stop off ground level, the animator could move it down "himself" and orient the soldiers feet so that they would remain parallel to the surface. The initial animation was done in Maya and cached animation clips were imported into Houdini as baked geometry. It was decided later to transmit the animated rigs to Houdini and do more work on them. This allowed us to complete the animation procedurally, based on the nuances of the terrain.
Animation clips were prepared in many different ways, some were manually animated, and some were made by using "classic" mocap with markers and infrared cameras. Others were taken from free libraries and parts were recorded using the Kinect. A few of these clips were recorded for the film "August. Eighth", but there was not time to use them in the project.
"I put the Kinect near my workstation, and played out the movements I needed, immediately transferring them to the soldiers." Alexander Lipilin, Animation Supervisor of the Main Road|Post, explains. "It was convenient. Guys at iPi Soft made the program, which was a unique solution for markerless mocap of the body with the use of Kinect. There is no equivalent of this technology yet".
Another subtask was the simulation of the clothes. Since the soldiers were dressed in overcoats and moved quickly the clothes could also be used as a tool to hide some flaws in the animation. The standard solver in Houdini was used for cloth simulation. Houdini was not the fastest compared to other solutions, but in this case it had one big advantage being a part of the studio’s pipeline allowing all the calculations to be done in parallel on the farm.
The film has one scene where we see soldiers on rafts and boats crossing a river. This scene also required a digital crowd, but it was more complicated than the one with exploding oil tanks due to the added element of filmed footage. The scene was already shot, but the live action was not enough and needed additional boats of soldiers placed among the filmed elements. They also had to be added in such a way that nobody could notice the difference between the filmed and the digital elements.
Water explosions also contributed to the complexity of this scene. In real life, an explosion in water raises a characteristic column of splashes but this can be difficult to recreate in a computer simulation. A brute-force approach explosion is spherical, so we needed to trick the computer. It was also important to find the initial speed of the splash expansion, and only after that could we form a column of water with the desired shape. It took a lot of time to solve this problem. Beautiful water explosions were made in the program Naiad, which proved to be excellent for a previous project "Metro".
Since the background in this scene was also replaced, as in most scenes of the film, we replaced all filmed water explosions with CG, as it would have been too complicated to rotoscope filmed water splashes in stereo.
One shot that proved especially difficult was where the soldiers float towards the shore out of the fog as all the fog effects had to be added in post-production. Originally it was planned to use compositing techniques for the placement of the fog, laying varying transparencies over areas with soldiers, but it soon became apparent that this would not work in stereo - even if the soldiers were cut out over rendered fog clouds. Based on the volume it became noticeable that the fog was placed behind the characters, not around them. As a result, each soldier had to be meticulously cut out by hand, and their depth in the scene had to be guessed by trial and error. This allowed us to gather the 3D information required to render out the fog with the plates of the soldiers cut out. Rendering was very slow, as an error of even half a pixel became immediately visible when the final application of fog was viewed in stereo, We spent a lot of time correcting edges on different characters.
In the film, the ruined city was not just a habitat for the characters, it was a character in itself, so it was treated with the utmost care. Initially, it was clear there would be a requirement to include a lot of shots including views of the city so Main Road|Post's CG Artists began creating the city using maps and historical surveys.
Modelers also had a lot of work to do, creating four versions of each house - unbroken, slightly broken, half destroyed and ruined. Later the artists developed a method to "destroy" house models in the procedural way using a dedicated algorithm (another interstudio development) that could be applied to an unbroken house to make it half-broken or ruined. The level of detail in the procedurally destroyed houses was not high so these models were used in the background and foreground buildings that were "destroyed" by hand.
It also became clear that all of the destruction needed to be covered with debris. Random sets of ready objects were modelled to create piles of rubbish which were then dropped on the ground. The results of the simulation were cached, then another layer was thrown on top. Several rubble piles were created this way and then artistically dumped around the city.
Trams, broken machines, even sunken ships at the pier were also placed throughout the city to ensure every city view resulted in an interesting image on the screen. Ironically, we effectively needed to breathe life into the war-torn city, and a lot of effort was spent doing it.
The lighting of the CG city required extensive work as it was intended to be historically and geographically accurate. One of the Main Road|Post artists who grew up in Volgograd knew the city well and would say "The sun rises here, and in the winter - there." But as it happened the correct geographical placement of the sun often meant it was behind the camera or behind a character and not where it would spectacularly light up the city or the river. Fedor Bondarchuk said during the first tests: "It's okay, but I need a river, it should be clearly visible. This is Stalingrad, the Volga is there." As a result, we decided to abandon historical authenticity in favour of artistic beauty.
In order for the city to 'live' and change a lot of small details required attention, events happened all the time - something was always exploding, blazing or fuming. While important, these small details made it difficult to work technically as the large and complex model was extremely resource-intensive.
"Every time something new appeared, we tried to optimize what was already existing." Stanislav Pologrudov, one of the key TD's in the studio explains: "It happened step-by-step, with new objects in the scene we were looking for new ways to optimize. Somehow we still managed to keep it within our capabilities each time, although, of course, calculations took many resources."
All the objects on the "map" were placed in Maya, with Houdini used only to place locator points with information about the building: its type, subtype, and stage of destruction - and so on for the rest of the objects. Before sending out renders, Houdini would take points from Maya with the encoded information and work over them in order to calculate, lay them out and read files from the disk with the certain model of a building and apply the textures assigned to it. All this was adjusted and implemented, with new objects and corrections being introduced constantly. The model of Stalingrad contained more than 13,000 houses and 6.7 billion polygons in total.
One of the most spectacular scenes in the film includes a dramatic sequence of a crashing German Heinkel He 111 bomber. The initial shot-by-shot breakdown required the opening of the scene to show the main character standing on a roof while the camera pans around behind him to show the city and the sky before catching first sight of the crashing bomber. The breakdown required the plane to brush against the roof of the theater, demolish the dome, hit the ground and come to a halt, at a given location. The final resting point of aircraft dictated the trajectory of the crash sequence.
A pre-visualization was done building the scene from beginning to end in the correct scale using a low-poly draft model of the plane. The pre-visualization was made to assist the DOP position the camera during shooting and doubled as the beginning of the animation work. It also now allowed us to place cameras and shoot the plane from any side. "We placed about 16-17 cameras to search for angles" VFX Art Director Andrey Maximov. says, "The very first angle was set according to the exposure sheet, and then it was necessary to tell the story of the plane crash and understand which angles were needed and how many shots could be made in sequence. We shot from 15 cameras, made cuts, watched, and then called in the Cinematographer Maxim Osadchy. He sat with us and offered his artistic decisions. After two weeks of work with Osadchy we decided on four angles, where at least one was fully CG".
Motion schemes of the camera were made so that everything would be filmed correctly on set. However, on the shoot day the correct camera crane did not arrive, and the one that was on set couldn't get the story boarded shot as the boom was too short. As a result, the filmed material differed greatly from that planned and significantly affected the look of the crash. The main problem being the stopping distance of the plane was reduced by almost half.
The scene was also quite difficult for camera tracking as the previsualization required the camera to lose sight of the plane, focus attention to the rotunda and the falling wing and then return to the plane, and the camera was constantly zooming. We decided with all this in mind a full CG shot would be cheaper and faster to make rather than dealing with all the problems associated with using a plate. This was also possible as the whole set had already been reproduced in CG and we thought it would not be necessary to model anything more specifically for this shot. The scene took six months to create.
In the final shot modellers had to make new geometry for the dome changing its size. The plane was also moved and the trajectory of its fall corrected. The main challenge with the shot however, was with the simulations as they had many different effects and numerous destructions which all proved to be very resource intensive. Rigid body dynamics were used for the visualization of destructions, then fluid solver was used to simulate dust. As the plane was large it also required deforming in flight with its wings needing to "tremble" in the collision with the building. These nuances were made with the use of soft body dynamics.
The scene also required fire and smoke (with displacement) to come from the burning plane which presented more challenges and three weeks of work. The amount of smoke in the scene was significant and a lot of it was close, requiring a high resolution, lots of memory and processing power. This required us to determine our available resources and optimize as much as possible. We also had to figure out how to simulate several clouds of smoke and seamlessly combine them. This is the moment when the camera view moved to the rotunda and losing sight of the plane really helped as it allowed us to switch the simulations, solving some of our resource issues.
The animation of the wing crashing into the rotunda was also a complicated task and there was always someone on the team who did not like the final animation. Finally it was decided to accept the laws of physics and fully simulate the destruction of the rotunda. The result exceeded our expectations.
The third shot in this scene was a slow motion shot of the plane flying past a window with characters in the foreground. In this instance the animation of the plane was very simple with only two keyframes, one at the beginning and one at the end of the scene. The complexity of this shot was in volumetric simulations, which were not easy to make in slow motion and needed to be heavily tweaked. The model of the plane "weighed" about 10 million polygons, as it was highly detailed with even the cabin being partly built "furnished” with controls. About 20 diffuse maps, a dozen specular maps, and several bump maps were created and in total about 40 textures in 4K resolution were made on the plane. There was even a pilot, although while watching the film he can hardly be seen. In the process of work the artists decided that the plane felt empty without a pilot inside and one needed to be added as the model looked unconvincing at "subconscious" level without one.
The most difficult part of this scene was the final part of the crash sequence. It took two months to perfect the way the plane interacted with the ground particularly simulating how it spread particles with its body or tossed them with its propellers as it hit the ground. In total this scene contained about 50 volumetric simulations (smoke, dust, contact with the ground), and about 30 rigid body simulations. To create a sequence it took around 30 to 50 layers of compositing, depending on the complexity of the shot.
Another challenging shot was the moment where the action follows the flight of a tank shell (before it explodes) into a room occupied by the main characters. The scene includes a complicated camera move that makes a quick track into a building through the set, slows down, breaks a door, and speeding up, flies across the room past the characters before exploding. All this needed to be run in slow motion before a blast wave destroys the furniture in the room and knocks the door out, scattering the characters. Finally, the camera pulls back, and gradually becomes engulfed in smoke. The creation of this shot took about three months.
The Director's vision was, to see the characters flying away from the blast wave. This was initially shot in live action using wire rigs. However when the plate was inserted into the shot at the animatic stage it looked very forced and grotesque so another shot with alternate choreography was needed to get the right effect. As there were not enough resources for a reshoot or to recreate the shot with CG characters, it was decided to remove characters completely. The Director agreed with this decision.
The peculiarity of this scene was that it was non-linear and slow-motion was included requiring action during the shot to accelerate and decelerate several times. When it comes to simulations, retiming becomes very problematic so it was necessary to come up with technical solution that would allow for quickly retimed simulations. Most simulations in the scene were solid state. To decelerate fragments of furniture and parts of the building inter-frame interpolation was used and volumetrics were simulated in the "slowest" time, so they could be accelerated later.
After creation of very precise animatics, Modelers built a detailed model of the room including all the furniture and objects, vases with flowers, books, and pictures on the walls. The creation of "wood chips", which fly away from the door in the foreground as the shell broke through it, was a technically complex aspect of the shot. It was necessary to show that it was not just the partition under the Voronoi’s formula, because wood actually breaks differently, by fibres. If this was done the wrong way, it would be noticed by even a casual observer so a specific pattern of chips and protruding wood fibres was needed It was also necessary to make sure that chips did not fly too deep into the auditorium. The scene was fine-tuned for long time in stereo, firstly in a three-dimensional scene, and then in compositing.
As all the furniture in the room had to be destroyed the choice was made not to favour physically correct destructions, but to construct shots based more on artistic beauty. This meant we could not destroy the entire room with one simulation but had to break all the objects into groups with each group of objects being simulated separately. For example, the team would simulate the destruction of a table with objects on it before moving onto the next group. Then the general wave of destructions were mounted from the elements, creating the overall dynamics and the rhythm of the shot. In general, about twenty rigid body simulations and about ten volumetric simulations were employed in the destruction of the room.
Another vivid scene of the film is a shot in the theatre, where a young soldier takes a girl to save her before the final battle. In the scene they look at an air battle through a hole in the theatre wall. A lot of work was done with the matte painting of the sky for these frames. Using photos of lightning as a base, a painting of the explosions in the sky was created for the compositors to show light and choreograph the flashes in the battle. The air battle was almost entirely made by using matte-painting and compositing in the night sky. Planes and shots of anti-aircraft fire present in the shot were created separately. "The fact is that today's shells fly in a different way, faster" says Arman Yahin "and the references of that time did not survive. Through trial and error and dozens of versions we finally achieved the desired effect".
Before the heroes draw the fire on themselves, their house is surrounded by German tanks. There were only two physical tanks on the set, so we had to add in another seven and work with their changing positions within the scene.
The most difficult shot occurred as a result of a problem on the set at the beginning of the attack on the house. According to the script the tanks had to destroy the brick pillars of the fence. On the set the construction of each pillar included an adjacent hole where the bricks would fall allowing the tanks to continue travelling. A technical hitch with the second tank ended up with it becoming stuck during this action and having to remain stationary throughout the duration of the live action shot. As a second take was not possible so it was necessary to "catch" the tank, replace it with a CG tank, copy its movements, and complete the action as if he had not got stuck. This presented some time consuming challenges as the camera was moving around the tank for much of the shot it so it was very difficult to replace the real tank with a digital tank. The solution to this problem took a lot of time - different variations of the animatic were created. The artists did their best to combine the new tank with the filmed one, which had to be wiped off out of the shot despite it taking up almost half of the frame.
All of the digital tanks required the animation of secondary movements to look realistic in the scene. We needed to study the movement of the tanks on their springs, how their moving metal parts interacted as well as how the tower and barrel, responded to changes in the motion. All these secondary movements were scripted as simulations requiring the animation team to only control the tank’s overall travel though the scene.emissions.
In many scenes of the film and particularly in the scene with the tank attack, ash can be seen falling from the sky. It was decided that this would be created in-house as a separate asset and it became clear in the early stages of this work that the team would need to be able to control the size of the particles, their behaviour, direction of movement and intensity.
Again, difficulty arose due to the film being shot in stereo as the insertion of a flat plate of ash would be immediately noticeable. To resolve this we decided to break each ash into several layers restoring the foreground, and inserting plates into the middle ground. Objects at different depths were used as the masks for different layers of ash, and this created the illusion of volume. Numerous smokes haze elements were also added.
In the final scene of the film shots of the destruction of the house take place from off-screen Russian artillery which, destroys utterly everything in the square.
Initially a, smokey and beautifully swirling TNT explosion was shot and our task was to recreate the filmed explosion and in less than a week a good result was achieved. It was like a real explosion in its dynamics and texture.
Then perfectionism came into play; everyone wanted to get a better and cooler result. They continued to work on the explosion, filled the remaining space with smoke, changed the light, and added the reaction of the ground to the shockwaves. We found these details to be very important as any explosion would look unconvincing if objects in scene did not respond to the shockwave and the movement of air. A tank jumps up because of a bump, dust rises, splinters fly in these frames.
"There is such a feature: a little change in the parameters of the simulation will lead to an unpredictable result." Alexei Zaytsev, one of the key TD's of the studio, tells. "And you will know about it only two days later, when the simulation is calculated. We had to operate very carefully and more intuitively with the explosion parameters, logic is of little help here. And if we change the lighting or camera angle, then the same explosion will look different again."
Clumps of earth, which tore parts of the smoke from its slender tendrils, were visible in the filmed explosions. However, in the computer simulated explosions, the stones and smoke moved independently. A solution was found, and everything was on schedule until we learned Director, Fedor Bondarchuk had a very different idea of how the explosions needed to look. "But they have to be absolutely different!" There was two weeks left to film delivery!
Pyrotechnic explosions from the films about the Second World War are quite specific, the explosions are mainly comprised of earth and need to show clods of earth flying from the thickness of the dirt and they must interact with smoke.
To create beautiful explosions on the film set, bags filled with different types of sand, stones, debris, chips and peat were put over the explosives. Then during the explosion the elements rose to different heights, creating patterns and refraction of light filling all the filming space. Together, with the background and the response of the earth to the shockwave, these explosions look very convincing. We were required to replicate this. A feverish search for solutions began.
"I understood what they wanted from me, but I also realized that it was not even a matter of time, but a matter of pure luck or chance whether I would get it or not." says Alexey Zaitsev "The dynamics of physical phenomena is always a roulette game. I can never tell when I can give the desired result with guarantee, whether I will be able to find the thread, which I will then pull, and everything will be beautiful or I will not be able to do that. You can search for a long time, or accidentally get the result in one day. There is no precise prescription as in cookbooks. These are all experiments, the trial and error method, guesswork. I knew, if now I could not find what was needed, we would have to admit that we were not able to solve the problem. And we did not want to admit it, because we had been working for a long time and with success. So I worked on adrenaline, set up flight speed of the clods, the force of gravity, but it was clear that we would not be able to go without the interaction of smoke and stones. And there was less and less time left."
“As a result, the smoke was divided into several parts, the first of which reacted with the clods of earth. There is a second layer of smoke, heavy one, it looks like sand that crumbles before smoke disperses. This was achieved by modifying gravity for volumetrics. And this effect completed reconstruction of the desired illusion. Finally Fedor Bondarchuk said, "Fine! This is what we need!" Days were left and a total of 55 versions of the explosion were made.
Digital characters also needed to be seen beside explosions. There was already a new approach used for CG crowd scene creation, six months had passed since work on the previous scenes with crowds, and it was decided to revise the entire technology. As a result, the studio completely rejected the use of Maya for animation, transferring the whole volume of work with the digital crowd scenes to Houdini. This strategy paid off, it took a long time to transfer elements from Maya in Houdini and as time was of the essence, the new approach assisted us to keep to the schedule.
Another solution was a complete rejection of manual animation in the crowd scenes as the studio relied on mocap for everything involving the animating of people up until now. The German soldiers had to respond to the shockwave from the explosions and it was going to take a long time to animate everything manually by frames so "ragdolls" came to the rescue. At the moment of the explosion characters who in the “lethal zone” of artillery shell explosions were turned into "ragdolls" and went to a rigid body simulation, which was rendered almost in real time. Then the result of the simulation was transferred back to the animated rig.
"It's like a computer game: you set up explosions, launch the soldiers, and at once you can see how they fly away." Alexander Lipilin explains "You look at the result, change parameters of explosions and turn on them again, until you get a suitable choreography."
One of the biggest challenges of the "Stalingrad" project was that it was in stereo and it was the first one of such scale to be undertaken by the Main Road|Post team. A lot of testing and learning was required. It was also the first project where concepts were made for all the shots before work started. An incredible amount of concepts were made so that everyone knew how every shot should look like in the end.
"Compositing in stereo is very different from usual compositing, because quality requirements increase.", Lead Compositor Sergey Shuppo says "Those things, like small flaws, which would be unnoticeable in usual format, in stereo are very visible. Any defect is noticeable to the eye as it is easy to see, for instance if the mask is curved or there is an extra glare. It is noticeable especially if this defect is in one eye only, our brain registers that something is wrong at the forefront, it is painful and unpleasant. That’s why it was necessary to observe errors carefully and correct them at an early stage, otherwise errors would accumulate, and it would not be possible to comfortably watch the film".
The requirements for the work was very high for the team as all work was checked on the monitor, TV and big screen. The bigger the screen, the greater the distance between the objects. Errors that cannot be seen on the monitor, like the object being 1-2 pixels out, turn into something with a depth much greater on the big screen.
A casual viewer will not understand what the problem is, but will experience unpleasant effects from the errors like headache, eye pain, or nausea. This is because the errors are perceived subconsciously and when a number of errors accumulate the brain collects them and conveys a feeling of discomfort to the viewer.
Our workload was huge. Work like cutting masks for rotoscoping required special care. In stereo these masks must be done very carefully, because discrepancies, not just in pixels, but in fractions of a pixel will be noticeable. When working in stereo it is very important to put objects at their correct depth, so in the final frame a background object does not end up in the foreground, or vice versa. In addition, the space in the depth of the frame needs to be correctly filled, otherwise the viewer will see emptiness. It is necessary to create the illusion of a single space, with the whole image in terms of depth, and not just arrange the objects in "layers".
Tracking and correction was all done manually. We faced new challenges at every turn, and learned to deal with them. For example, how to treat chromakey at the foreground if there is fire and warm warm air distorting the picture differently for stereo.
At the beginning of the work on the project there were only two people who have had experience with stereo in the studio. During shooting three employees were assigned to be on the film set at all times, and they were rotated, so ultimately 12 people gained invaluable experience.
"This was an extremely difficult project.", VFX Supervisor Arman Yahin sums up. "There was an incredible amount of Herculean effort invested both into concepts and modelling, and into the development of explosions, and work with stereo".
Some of the rotoscoping was outsourced to a group of freelancers from Belorussia under the direction of Alexei Dubko as the team had previously worked with Main Road|Post on "August. Eighth." During production the Belorussia team started another project and a second contractor had to be found. The work was continued by Kiev studio, Braindraintrain. They had not worked with stereo before, but with the help of the compositors from Main Road|Post they quickly learned all the nuances and completed a lot of shots.
More than one hundred compositing shots were undertaken by the studio Pixel Bears. They added tracers and bloody splashes into the fight scenes. They also assisted with shots that needed refinement or sky replacement, and shots with sniper scopes. Other rotoscoping work that was outsourced consisted of the shots for final tank attack scene, where three-dimensional tanks and other CG elements needed to be placed behind real objects and where CG ash was added. There was an extensive amount of masking required.
It was very pleasant to work with Fedor Bondarchuk, he expresses his wishes very clearly, makes decisions quickly and was happy to adopt new technologies. The team encoded dailies for iPad or iPhone, and sent them via e-mail. He watched and commented on everything. The work was very constructive and was productive".
Stalingrad will be released in a further 45 Countries in February 2014.
Directed by Fedor Bondarchuk
Produced by Sergey Melkumov, Anton Zlatopolsky, Dmitriy Rudovsky, Alexander Rodnyansky
Written by Ilya Tilkin, Sergey Snezhkin
Music by Angelo Badalamenti
Principal Photography Maksim Osadchiy
Editing by Igor Litoninsky
Starring Petr Fedorov, Thomas Kretschmann, Yanina Studilina, Dmitriy Lysenkov, Aleksey Barabash, Andrey Smolyakov,