The bots are back in Transformers: Revenge of the Fallen, and ILM tells Bill Desowitz how it rose to the challenge of animating and getting superior performances.
You just knew that Michael Bay was going to escalate everything for Transformers: Revenge of the Fallen -- and he has. There are 46 robots in the sequel compared to 14 in the previous movie, and some of the new ones are bigger and more complicated, with a lot more components and weaponry. In addition, in trying to raise the IMAX stakes after The Dark Knight, Bay decided to shoot the two primary fight sequences using the large format camera, posing more challenges for asset management and rendering.
Of course, Industrial Light & Magic was up for the challenge for doing another Transformers, creating a lot more work in the same amount of time, with Scott Farrar and Scott Benza once again assuming their roles as visual effect supervisor and animation director, respectively. We also spoke with Alex Jaeger, who reprised his role as vfx art director.
"There were a lot of comments from people on the first one that they enjoyed the transformations, so we knew we needed a certain amount of plussed out, easy to read transformations," suggests Farrar, who was also second unit director. "And we knew we had a lot more robots after reading the script -- four times the number as the first movie -- in the same amount of time: about a year-and-a-half. So that was a challenge. It just meant that you have to start the design of each robot sooner. The average robot takes about 12 weeks to build the model and then about another 12-15 weeks to do all of the rigging and, of course, the painting. So we had a huge model staff, we had a huge animation staff; everything was bigger on this film to proportionally get the work done."
Farrar adds that when you factor in all of the assets, ILM reached 60 major builds on Revenge of the Fallen. (Digital Domain worked on a few of the new bots as well, including Alice, who resembles an X-Men mutant.) A favorite factoid: disk storage space was up 150 terabytes from 20 terabytes on the first film. And at capacity they used 95% of the render farm. And the crew hit 350 for the last couple of months.
"One thing that was a major breakthrough was allowing the animators to do most of the motions," Farrar continues. "A big problem for us, up to this point, has been they do some animation, we see it in the shot, the creature dev people, who manage all the pieces on the character, might make a gear move; then they would send it back to animation to plus it out for us. And this back and forth is very tedious. So we arranged everything so there is more control in the hands of the animators. That saved weeks. Another great time savor was the ability to break apart the lighting on each of the robots into various categories (specularity, diffusion). You can render the whole robot with everything baked in, but then you have no control over it later. If you want to fluctuate lights and darks and control that in the hands of the compositors, we can do that with a little bit more disc space, but it gives you room to control the image at a really critical point when Michael wants last minute changes in the lighting.
"And we tried to make a lot more efficiencies because this is really tough work on everyone. Our simulations got better (but not smaller). This is the largest simulation project ever at ILM (when Devastator is breaking apart the top of the pyramid). It's colossal! And Devastator is colossal! He has 80,000 parts compared to little bitty Optimus, who's got 10,000 pieces. And Devastator in real world terms is about 150 feet high. And he was in IMAX shots, which are even bigger."
Adds Benza: "We had to raise the bar and to do that Michael's idea was to hit the audience with a greater number of robots and robot types with varying personalities. And we were able to bring a higher level of quality as far as the performances and in the fluidity of their movements."
"This one was about letting the animators do more," Jaeger suggests. "My part was making sure these things have all the detail they need since we have a lot of new characters, making sure that they fit in with our old characters as far as detail and designing all the damage for the robots: keeping together all of the reference -- this part should look like bent metal, this part should look like glass. And not only for the robots but also for the aircraft carrier sequence. That was a huge deal coming up with the look of it and technically all of the pieces involved -- hundreds of thousands of pieces.
"As far as the design, I came up with a lot of new weapons because we had a lot of new Decepticons. Even some of the Autobots. Everything from grappling hooks to guns to rockets.
"The intention with The Fallen [the head Decepticon rival] was to make him more organic to better differentiate him from the others. Another new thing: both Michael Bay and Scott Farrar wanted to introduce the fact that these characters are alive. Drooling and spitting and bleeding and breathing. Rather than mechanical beasts standing around."
Farrar also touts the higher degree of acting. Jetfire, the old SR70, is the best example. They added more pieces to the face as they went along to increase articulation and emotion and line delivery. "He has a sneer at one point, so we had to redesign the face and the eye area so he could wince," Farrar explains. "That's the new, big challenge: to show more emotion, which was our main goal. Cars are the hardest things to light and each one has to reflect the environment. And there were a lot of environments. Ravage, the stainless steel cat, was tricky. He's a heated blue, darkish steel. You spend a lot of time fishing around for what the look is. You light it, you look at it and you keep add some scorching, you keep fiddling with things until you finally have something that looks good."
Benza says they did a full series of tests for Jetfire. "It was hard to find a guy who had a grizzled, veteran feel to him but was also decrepit or broken down," according to Benza. "Through a series of reference we found [the right mix] of acting and voice. We went down the path of doing a compilation and it ends up being what you see in the film. He's a war veteran that's tired of fighting for the wrong side. 'Too old for this shit' is how we described it and that line actually made it into the movie."
Benza offers that they greatly improved the fight choreography. "In the first film, the fights were much shorter and less choreographed. We put a lot of time and effort in fleshing those out. We utilized a lot of martial arts influences and spent about two months doing previs for forest fight sequence and did a full three-minute animatic (in Maya) before the plates were shot for that sequence.
"The only requirement from Michael was he wanted to see hooks coming out of Optimus' hands and rip the helicopter robot's face in two. That was the one idea that we wrapped the entire fight around. The fight is a couple of minutes long and a longer version of the fight sequence appears in the IMAX version.
Interestingly, Benza's personal contribution was pretty unusual. "I did a couple of editorial passes on the two fight scenes: the forest fight and the fight between Bumblebee and Rampage. Since we had done previs, I was on the New Mexico shoot, where we did background plates for that scene and we had another representative that supervised the shoot of the Bumblebee fight, and Michael thought it would be a good idea to basically dump all of that footage in my lap to take a first pass at editing those fight sequences together. So I went through all the footage and found the best-looking plates for the previs that we had designed. Not everything was designed to match the previs, but I was able to come up with a proposal for him about how the sequences should be put together and gave that to the editors, who then made adjustments and refinements and condensed it for what they wanted with Michael. It was a pretty neat experience."
Meanwhile, shooting the forest fight and climactic battle in IMAX was great fun but a concern. "Devastator was going to fill the frame along with Digimatte backgrounds, which is all res'd up from stills, so it was all us in that frame, which is eight times more film frame than 35mm anamorphic frame, eight times more disk space and eight times longer to render," Farrar says. "It took 72 hours per frame to render Devastator. We tried a little bit of everything for the forest fight. It's still an action sequence but the hard part was shooting thematically. We know the IMAX people tell you to slow the camera down and lock it off. Well, that's not how Michael shoots. We have pause moments, where you see the characters slow down, but then we have high speed and go back and forth that way. It's always experimentation and I hope it's successful. It is a fight after all.
"Even there, we were doing little odd things, like variation percentage of motion blur on the characters. I tell everyone to jump on the Harryhausen band wagon -- that's what we've gotta do. People in computer graphics don't want to reduce motion blur, but the problem with the robots is that they have so many little pieces that they become artifacts with so many sharp things moving through the frame. I found it's better to reduce motion blur in certain moments, like when Bumblebee comes close or Starscream has moments in the forest fight and Optimus and Megatron, where we reduce motion blur to half, a third and an eighth."
Given Farrar's photographic background, he continues to push really hard for better lighting, shading and rendering. "Everybody knows that I'm trying to create a lighting package that emulates real world photography. We're not there yet but everyone has the hang of what I'm looking for stylistically. A good example is Jetfire. He's about 50 feet high and in computer graphics terms you can put a single light source out there and light the whole object. If you're out in the sun, that's one thing. But if you're indoors, like in the Smithsonian, that's not real in human terms. So what I like to remind everyone is that it's like lighting a real set, where you have a maximum of a 20K HMI. And that will cover so much of a range, but there's falloff in every area -- that's physics. And so you have to supplement. Still, you'll see imperfections. But the problem with computer graphics is that it's so perfect. So we try and induce these imperfections. So Jetfire in the Smithsonian required around 41 actual lights and we chose to make them smaller and human size. And we hope it will seem more real to the viewer."
There are a lot more environments in Revenge of the Fallen, allowing greater diversity for eye candy. The two main locations are Bethlehem Steel in Shanghai and the pyramids of Egypt.
"It was an honor to shoot in these places," Farrar admits. In Shanghai at night, that was a week of prep just to get all those lights up there. The biggest problem with our large robots is to show scale. And the pyramid backgrounds were fantastic for that. A key concern was we were not allowed to do aerial photography at the same time the crew was there. Permits were not approved when the main unit was shooting, but we had our Digitmatte crew there and our stills crew was assigned to get every side of every pyramid close and at distance. Our big worry was that we would not be approved to shoot aerial plates. So we were canvassing the area and it would've been really, really hard to recreate based on stills only. Fortunately, I think it was in December, we got approval for the aerial shots. But the stills gave us a lot of textures. Almost everything we do in computer graphics backgrounds I like to glean from photography as much as possible so it'll have a realism to it."
The other noteworthy development was that the breaking apart of a pyramid top was eight times bigger than the previous ILM rigid simulation record. It only required four or five shots but that took seven months just to create the simulation of the blocks tumbling and being torn apart by Devastator. TD Christopher Horvath oversaw the pyramid sim work.
"It has caused our simulation folks to improve everything and we've got some really clever people that understand cinematic movement better, not just the technical elements," Farrar says. "We push everybody to observe what is real."
Bill Desowitz is senior editor of AWN and VFXWorld.