Although the slick visual effects in DreamWorks’ figure-skating flick Blades of Glory easily slid past audiences, the filmmakers for this box-office hit would have been skating on thin ice without the skill of Rainmaker Visual Effects. Rainmaker’s Visual Effects Supervisor Mark Breakspear led the 150-person team that filled 3D stadiums with digital crowds and gave the four main actors Olympic-class skating moves — or rather, convinced the audience these skaters, not their stunt doubles, could perform an Olympian’s triple salchows, double axels, spins, lifts, and spirals.
Click the link below to see a Flash presentation (broadband recommended) featuring Rainmaker's work on the film, including videos and before-and-after slides.
Directed by Will Speck and Josh Gordon, the film stars Will Ferrell and Jon Heder as an unlikely male figure-skating pair. Writes Kevin Crust of the Los Angeles Times, "Whatever combination of choreography, camera trickery and special effects were required to render the over-the-top, hyper-real skate numbers, they're executed with wit and ingenuity.”
That wit and ingenuity was due, in part, to a lucky break. “Jon Heder cracked an ankle bone during practice, so we had a three-month hiatus while he healed,” says Breakspear. “That gave us an opportunity to previs every shot in the skating sequences, and we came up with new moves.”
To make the moves work, Rainmaker used new, state of the art techniques to replace the four stunt skaters’ faces with the actors’ faces. The techniques rely on a blend of commercial software and proprietary tools. “People have done fantastic face replacement in CG on mega-movies, but we didn’t have four years or a blockbuster budget,” Breakspear says. At first, the studio pushed for a 2D solution, but Breakspear convinced the directors that switching faces in 2D would limit their ability to shoot the skaters and restrict the skaters’ movements.
Fortunately, Rainmaker had a gold-medal face-replacement expert on its team, CG supervisor Kody Sabourin, who had worked on the legendary “Super Punch” shot in The Matrix: Revolutions. In that shot, accomplished with proprietary technology, Agent Smith’s face deforms in an extreme close-up when Neo lands a punch.
“Given my experience on the Matrix films and from working at ESC with people like George Borshukov and Kim Libreri, I knew the best way to approach this project was with animated textures that we captured at the same time we captured the performance,” he says.
Re-inventing Face Replacement
At first, Rainmaker tried to hire a studio or company to do the work. “We couldn’t find anyone we could outsource from,” Sabourin says. “People could capture the performance, but not the textures, or if they captured textures, the textures were low-res and not usable. Or, they’d texture-map a still frame on an animated mesh. I knew, based on my experience on Reloaded and Revolutions, that wouldn’t work.” As a result, Sabourin led a team at Rainmaker that developed the process used for Blades of Glory.
They began by taking digital photographs and making plaster casts of the actors’ faces. Then, they had the casts scanned in high resolution at XYZ RGB in Ottawa. Using the photographs and a 3D model created from the scan data in Autodesk’s Maya, visual effects artists determined where to place 102 tracking markers on each face. Before the performance capture session, makeup artists copied each marker’s position onto each actor’s face.
To capture the actors’ facial performance and animated textures simultaneously, Rainmaker devised a clever set up using three HD cameras, one film camera, and mirrors. “Imagine that the actor is sitting in a chair and around him are four cameras about two meters away,” says Breakspear. “Above and behind him are mirrors.”
The team captured the tracking markers with the three HD cameras, and recorded textures during the actors’ performances with the film camera, which they positioned in the center. Breakspear put lights behind silks to capture the faces with completely flat lighting. “Using the film camera was something that has never been done before,” Sabourin says. “Not even on the Matrix films. We could scan the film and have textures at 4K resolution if we needed, whereas with HD, we were capped at HD resolution.”
The actors performed to footage of his or her stunt double. “We had split screens on set so the actors could watch the stunt doubles in the plate and then watch themselves reacting,” says Sabourin. “The director had free reign.”
The mirrors doubled the data collection and provided textures from multiple angles. “With mirrors, rather than using more cameras, we could split one image into three,” says Sabourin. “We got textures from the front and both sides of the face, and an extra three cameras tracking the performance.” To match the frames once the cameras rolled, a time code generator linked all the cameras.
The effects team tracked the markers using RealViz’s MatchMover Pro and triangulated the actors’ facial motion into 3D space. “We ended up with 102 dots that looked somewhat like the actors’ faces moving in 3D space,” Sabourin says. “That is, you could make out many of the moving shapes, like their ears, throat, lips, eyes, and nose.”
Next, they applied the animation from the dots to animatable meshes created from the high resolution XYZ RGB scans of the actors’ facial casts. Proprietary code helped them transpose as much detail from the actor’s performance as possible. “That gave us an animated face mesh,” Sabourin says.
Weight maps helped make that movement believable. “We painted weight maps – grayscale texture maps – to define range of movement,” says Breakspear. The weight maps controlled how much influence each dot had over the geometry. “The skin on the nose doesn’t move far, so the weight map would be tight there,” explains Breakspear. “But a weight map for the cheek would be squashy.” Thus, if an anomalous dot tried to stray from reality, the weight maps pinned it in place.
The result gave them fully CG versions of the actors’ faces performing as they did on set. “It was a breakthrough,” says Breakspear of the team’s solution. “A eureka moment.”
Getting the Texture
After a final cleanup of the performance using hands-on animation — a tweak to animate the lips or the eyes a bit more, perhaps — the effects artists began texturing the faces. “We had textures captured at 24 frames per second that matched the performance we captured,” says Sabourin, “So we applied those animated textures to the mesh.” To tidy the textures and remove the tracking markers, the artists used Eyeon Software’s Digital Fusion, Right Hemisphere’s Deep Paint, and Adobe’s Photoshop.
Then, especially for close-up shots, they extracted high-resolution displacement maps from the XYZ RGB scans to add such details as scars and pores. “That, combined with the animated texture maps, gave us wrinkles,” says Sabourin. “Because we had 150 face replacements with unique performances to do, we gave the faces that were far away only the bare minimum, but a lot of the shots were close-ups.”
To match the lighting on the digital face to lights used when filming the skating doubles, Breakspear relied on HDRI. “For every shot of the doubles, we had shots of a gray ball, a chrome ball, and images taken with a fish eye camera,” he says. Even so, the lighting team faced unique challenges. “The ice created a big bouncing light environment. Our lighting team did a fantastic job making the digital faces look right.” For rendering, Rainmaker used Mental Ray.
Crowd Control
In addition to the facelifts for the skaters, Rainmaker also did digital surgery on the LA sports arena where the directors shot all live-action footage, but all the action in the film takes place in three other skating rinks in three different locations.
“Creating full CG stadium replacements with CG crowds was roughly half our work,” says Breakspear. Although the studio had devised a method to create crowds in 2D for the film She’s the Man [see F&V's coverage here], the camera moves needed for the swirling skating sequences dictated a 3D solution.
During the live-action shoot, 800 people and 5000 inflatable stuffies sat in the stands. “The stuffies worked well for a blurring background,” says Breakspear. “They look like real people. But in some shots we needed to have the people moving and clapping.”
For those shots, Rainmaker either replaced everything in the L.A. arena from the second level of seats up or created fully 3D stadiums by building digital sets in Maya. For the crowds, the studio used Massive software, building brains that signaled the fans to clap, cheer, stand up, sit down, walk through the aisles, and even order hot dogs. The CG characters moved using animation cycles created from motion data captured at the studio’s new division, Mainframe Entertainment, which Rainmaker acquired during 2006.
Beware of the 'Iron Lotus'
The climax of the film revolves around the “Iron Lotus,” a secret move never before successfully performed, that the characters played by Farrell and Heder try in front of a crowd of 15,000 screaming fans. “It has never been performed before because if a skater gets it one percent wrong, it could decapitate him,” says Breakspear. “We dreamed up this move during previz. When we explained it to our professional skating doubles, they wouldn’t speak to me for a week. They were teasing me. We all had a good laugh making it.”
Creating the Iron Lotus moves for the film involved face replacements, rig removals, CG crowds, CG stadiums, and green-screen composites. “Everything came together,” says Breakspear. “It was so dramatic. John Heder grabs Will’s leg and throws him in the air. Will spins in the air backwards. Then, John puts one leg in the air, does his own spin and jumps into the air ... and ... it’s so dramatic.”
That people walk out of the theater laughing, not wondering how the actors could possibly have performed their skating moves is a tribute to the VFX moves performed at Rainmaker. “Without our new face-replacement technology, we couldn’t have done it,” says Breakspear.
VFX Supervisor Mark Breakspear
Facial performances were motion-captured — along with film-resolution textures — as actors watched their skate doubles perform.
Using scans of plaster casts, the filmed textures, and the motion-capture data, animated face meshes were created.
Images and video (C) 2007 Dreamworks LLC. All rights reserved. Images courtesy of Rainmaker.
Comments (10) for "Cutting-Edge Face Replacement ... on Ice"
1.
I'm a little confused why you had to go through the process of making a plaster cast. Couldn't the actual faces be scanned to make the 3D model? What am I missing here?
Posted by George H. on Tuesday, April 10, 2007 @ 09:12 AM
2.
Mark, nice digital stuntwork. These face replacing techniques could change a lot of things about moviemaking and postproduction. What other kinds of movie scenes do you see applying this kind of technique to?
Posted by Laszlo on Tuesday, April 10, 2007 @ 10:37 AM
3.
To date -- and I’d love to be convinced otherwise -- there doesn’t seem to be a system out there for capturing the detail in the face that a scan system like “XYZ RGB” can do. You get a little bit of the subtle detail from the handheld scanners, but the combination of subject and camera movement makes the scan miss the pore details or the tiny creases that come from the side of the main crease near the eyes. Having a solid, unmoving plaster cast is in my opinion by far the best way to get that level of detail. I’ve used both and, depending on the feature, a lesser detailed model might have worked for other types of facial effects. We needed to make sure that the detail was there from the start as doing full CG faces that are acting up close to camera is an unexplored country and we didn’t want to go into this sacrificing details from the start. Like I say, if you’re a 3D scanning vendor, please show me examples of the maximum level of raw detail that you can capture and I’ll be more than happy to rethink my opinions. The other thing is that the actors normally find it quite exciting to have a cast made from their faces and always want a copy for their mantlepiece above the fire. Creepy, but understandable! As for cost, I’ve been told the plaster cast way is too involved and costs more... Maybe up front you spend a little more time getting the quality you want, and the cost is a very small % more, but then you save a LOT more at the back end when you have a model that has all the detail you need in it and doesn’t need to be worked on for weeks just to get it ready for what you need.
Posted by Mark Breakspear on Thursday, April 12, 2007 @ 04:57 PM
4.
This is a great question and it cuts right to the heart of what made Blades so exciting to work on. Essentially any actor can now “do” the stunt in a much more visually exciting way. A spoof of any Olympic sport could have any actor diving off the highest board, or pole-vaulting to victory. Blades was just the tip of the iceberg. How people take this technology and move it on -- that’s something I’m very excited about.
Posted by Mark Breakspear on Thursday, April 12, 2007 @ 05:19 PM
5.
Didn't the soft omni-directional lighting of the ice rink actually make the job of face replacement a lot easier? If the stunt double were outside in bright sunlight, a 180-degree turn of the head would be a nightmare for the animators...right?
Also: how did you handle the problem of hair from the wigs flying in front of the faces?
Posted by David Halliburton on Friday, April 13, 2007 @ 08:12 AM
6.
Look all I want to see is the video you are offering. Why is that so hard?
Posted by robert dalva on Friday, April 13, 2007 @ 10:33 AM
7.
David: Yes and no. We shot the actor performances in a multi-directional lighting environment in order to remove any shadows on their faces. We did this so we could light them in CG to match the scene. The presumption that the ice rink was a soft lit environment is an obvious thing to think, but it's unfortunately not the case. At the LA sports arena you have a back of very bright lights on the north and south of the stadium, you then have the lights that we added for each shot on a case-by-case basis. Even with the bounce off of the ice, and those lights, you still had some pretty complex shadows on the faces to match. We gathered HDRI reference for every shot so our animators could match the on-set lighting quiet easily. If anything, I think replacing faces on strongly lit subjects would be easier, as your brain sees the higher contrast and for one reason or another accepts it as real a lot easier than if it sees a subtle difference in shades on a more low-contrast subject. As with most of these things, it's probably a throwback to when we would need to identify a well-hidden predator that is creeping up on us. Sheer survival instincts have made doing visual effects all the more difficult!
As for the hair, we developed a few good techniques for extracting the detail from the doubles head through a combinationn of roto and difference keying. We did a lot of both!
Posted by Mark Breakspear on Saturday, April 14, 2007 @ 07:14 PM
8.
its good, but the face replacement its need a real touch of
naturalism for lighting and color scheme.
Posted by jojiemcasiao on Sunday, May 20, 2007 @ 08:35 AM
9.
Dear Sir/Madam
I am wondering whether you have any video software that enables the user to change the face of actors with anothers faces of my choice
Posted by CHADI on Tuesday, December 2, 2008 @ 03:11 PM
10.
It's OK for safety reasons, and just to get the job done on some shots.
But you can't have everything just becoming fake, like doubles who walk, stand, move or dance with more poise than the actor. Performance is a part of the aura of some movies, like Fred Astaire and Gene Kelly movies.
Audiences should expect and demand real performance where it matters.
Posted by PT on Thursday, March 12, 2009 @ 03:38 PM