Studio R&D director Paul Carmen DiLorenzo walks us through the studio’s groundbreaking and innovative production system being recognized at this Saturday’s MPAA Scientific and Technical Awards.
One of the many achievements being honored at this Saturday’s annual Motion Picture Academy Scientific and Technical Awards is DreamWorks’ Premo animation platform.
Initiated in 2008, first rolled out on production during How to Train Your Dragon 2 and used on all subsequent studio-produced feature films including the current Oscar-nominated Boss Baby, Premo was designed from scratch to take advantage of the then relatively new multiple core CPU technology through the scalable use of parallel processing. DreamWorks engineers were able to harness the full computational power of multi-core HP workstations, providing animators a real-time creative tool that not only introduced more intuitive and artist-driven workflows into the animation pipeline, but eliminated numerous time-consuming processes where animators would waste multiple minutes, over and over each day, waiting for the execution of simple processes like viewing shot changes or loading a new set of character assets for a scene.
For their key contributions to Premo’s development, the award, which recognizes individuals who have added value to the filmmaking process, will be presented to Alex Powell for the design and engineering, Jason Reisig for the interaction design, and Martin Watt and Alex Wells for the high-performance execution engine of the Premo character animation system. Each will receive Academy certificates in the Scientific and Engineering Awards category.
AWN spoke with Paul Carmen DiLorenzo, DreamWorks’ R&D director overseeing Animation, Rigging, FX, and CFX, who explained Premo’s main features, how the system improves the creative process and enables a huge jump in production efficiency, and the challenges the studio faced building and implementing a software system that fundamentally changed the entire way they produced animated films. Please note, a complete list of contributors DreamWorks wishes to thank is provided at the end of this interview. Read the full Q&A below:
AWN: What are the practical animation challenges that Premo addresses? How does it make the animation production process more efficient and how does it enable and enhance artist creativity?
Paul Carmen DiLorenzo: Three main things. The first thing is character performance. In the past, with our old software, Emo, when you were posing a character, you don't see the actual fully deforming hi-res geometry. You’d see stand-ins…they're low res or punched out cylinders that approximate what the geometry would look like. The system does it that way because there are fewer computations [for the workstation CPU] and the animator can view close to real-time posing performance. But, that view is only an approximation that kind of gets them in the ballpark. To see a higher resolution version of their animation, with the full rig and computations, they would have to switch to what we called a “recalc.” That would create a movie they could view of the whole shot. It wasn’t rendering. It was just doing a better deformation of the character for the whole shot. They would look at the movie, see what the motion looked like, go back to changing the stand-in, edit some more…it was a really clunky workflow.
It wasn’t very intuitive. It could also take a while. They'd go get a cup of coffee…it really broke their process and efficiency.
AWN: That meant they constantly broke their concentration and wasted a lot of time over and over again all the time.
PCD: Yes. For example, if they wanted to test something with a character, it would take hours. So, that’s an issue of performance. Premo refreshes in real-time, providing high-resolution fully deforming characters and environments that better resemble the final look. Immediately onscreen. The second thing is much better fidelity, such as with the characters. Like I was mentioning about the deforming geometry. Switching back and forth [to generate a higher resolution deformation] was bothersome. There was also the fidelity of the scene or the shot around it…the environment and number of characters. A lot of times, we could only work on two or three characters [at a time]. But if there was a big scene with more interactions, you had to visualize how what that all needed to look like. That was really tough on the animators. Do two or three characters, figure out what you want those to do, then work on another two or three characters. As the scene complexity increases, the efficiency drops.
These issues are all tied together with Premo in better, more fluid workflows. You have workflows around posing and playback but also drawing. Everything is happening in real-time. Premo also has great integration of an animation library, which is a global library of poses for each of the characters. Supervising animators will setup a whole bunch of face and body poses -- smile, frown, mad, sad. Now an animator can double-click on it [a pose], it will set the character in that pose and then they can tweak it from there. It's very important that animators stay on model, that they don’t pull on something, like a cheek, and break the model.
AWN: It’s obvious that real-time refreshing alone would have a hugely positive impact on productivity and efficiency, not having to constantly wait around while a scene is processed. Did you ever perform any quantitative assessments of just how much more productive Premo made your artists?
AWN: Wow. That’s a significant increase.
PCD: They can push out two or three times more footage than before. I'm talking about the top animators. For more junior animators, they're probably pushing close to maybe one-and-a-half or two times more. So overall, animators are maybe pushing double their footage.
One thing to keep in mind though is that animating and working with Premo is not their whole job. So, it's not like the film can be done in half the time. But, when they're just sitting down and animating, I would say, Premo has easily doubled efficiency.
AWN: That’s tremendous. With that increase in efficiency, the ability to iterate more and work a scene more thoroughly, from a creative standpoint, what does Premo allow an animator to do better than they were able to do before?
PCD: That's a good question. A lot of it has to do with iterations. The core workflow of pose, scrub [manually scrolling through the animation, backwards and forwards, to preview the movement], pose, scrub, has now been refined to where they're able to animate as quickly as they can think. So, they can explore new ideas a lot more quickly and efficiently. “I wanna try these two characters doing this. Oh, that didn't work. Let me try this. Oh, that didn't work. Let me try this. Oh, that looks really good.”
The other dimension it impacts is the subtlety of motion. Once they get it [animation they like], they can iterate it. Work it. Refine it. You know, the subtlety of maybe a little eye movement, a little lip movement or a head tilt. It’s the difference between when you’ve been on the shot for so many weeks and when you're kind of just done versus you've been on it for a couple days and now you're really refining it.
AWN: From a systems development standpoint, how does a highly technical team of developers work in tandem with a team of artists to design, develop, test and roll-out such a major piece of technology? How did you make that collaboration work?
PCD: One of the advantages we have at DreamWorks in building proprietary software is we sit literally on the same floor as the animators. We work very closely with them and try to build exactly what they’re looking for. We work directly with a “product owner,” like Fredrik Nilsson [Premo product owner and animation workflow director], who’s coming to the table with 20 years of experience as an animator. He talks to animators on the floor and understands what they're looking for. We worked closely with Jason Reisig, who was the HOCA [head of character animation] on Home. He was kind of the product owner at that point, giving us an idea of the animators’ vision. Then we would figure out what technology we needed to build.
You touched on an interesting point. When Premo came out, it was a big shift, a massive shift. When we go through this type of change, having an animator integrated, assisting us with that change management, helps a lot. That person is talking to people that have the same background. They can help explain a lot of this stuff like, “Look, try this out. Look at how much better this is going to be. Look at this, look at that.” So, having a person who is going to put the system on their show, explain how the change is a positive change, how this is going to be really great for the animators, helps out a lot.
When technologists say, “Look we came up with this stuff and we don't really know animation, but this is gonna be great for you,” it just doesn't go over as well as someone who has worked with us on the development for several years and tells the animators, “No, look, this is gonna be great. I've tried it, I've been working with the developers. This is amazing stuff.” We also did what we call “milk and cookies,” where we have bi-weekly department meetings where we would bring in all the animators and show them our progress.”
AWN: When did work begin on Premo and how long did it take to develop?
PCD: Initial discussions started in 2008. The real inflection point was when we realized that CPUs weren't getting faster, they were just getting more cores. Underneath Premo is an engine called LibEE, that can do both graph and node parallelism [parallel processing across multiple CPU core units]. So, it's built to scale as far as the number of cores. Emo was not able to scale. It was built for single cores. Back in the day when CPUs were getting faster, 1 gigahertz, 2 gigahertz, you’d upgrade to pick up speed. I think it was around 2005-06 Intel started coming out with multi-core processors and that changed the game.
AWN: You had to build new software programs to take advantage of increased processing power.
PCD: Yes. Emo was getting old. So, the conversation was also, “Well, we should be building a new animation tool.” We took Emo as far as we could. The decision was made around 2008 to build a new software tool. It was first deployed on How to Train Your Dragon 2 in 2012. It took about four years to develop and maybe a year to fully deploy. Since then, we've continued to improve it. One of the big improvements we made in Premo 2.0 for Home was called “multi shot.” Before, to go from shot to shot, you had to shut down Premo and bring it back up, which took maybe three or four minutes. You had to reload all the characters and assets of the shot you wanted to work on. But, we figured out how switch from shot to shot. It took three, four minutes down to seconds. That also allowed animators to see the other shots around their shot, pulling from media services [DreamWorks’ proprietary cloud-based digital design and media platform that Premo integrates with]. They could see the continuity of their shot and work across edits.
AWN: Looking back, what was the biggest challenge you faced moving your entire animation pipeline to a brand-new technology infrastructure?
PCD: For me, there were three main challenges. First, there was the artistic challenge, for our artists and animators, of figuring out what is the best way to animate. Rethinking animation, that's a huge challenge. I've been doing it this way, just blow it all away and now the sky's the limit. How would you love to animate? What are the real problems? What is holding you back? That was a big challenge for the artists.
Then there was the big challenge on the technical side of building for multi-core. At that time, it was a fairly new thing. How do we do graph parallelism, how do we build the system and the application so it can scale, be flexible and last 10-20 years?
And finally, there was the management challenge. This is a massive program. How do you get all these engineers and animators working together? For studio management, getting all of these groups going in the same direction, the right direction, was a big challenge as well.
In addition to those honored by the Academy, DreamWorks would like to specifically congratulate and recognize the many contributors over the years who have made Premo the industry leading platform it is today. They are:
Adriana Damiani, Agustin Perez Paladini, Alejandro Boada, Alejandro Mier, Alex Powell, Alex Wells, Alvin Tenpo, Andrea Parkhill, Andrew Van Pernis, Andy Brown, Anson Chu, Anthony Lani, Ares Devaux, Atul Patil, Barathy Rangarajan, Ben H. Kwa, Bob Morgan, Brendan Duncan, Bridgette Wiley, Bruce Wilson, Cara Malek, Carlos Mellano, Carmen Badea, Cassidy Curtis, Chi-Wei Tseng, Chloe Chao, Christina De Juan, Colin Smith, Cyrus Wilson, Dale Alan Hoyt, Dane Stogner, Daniel Dawson, Daniel Muñoz Trejo, Danny Hahn, David Lee, David Stover, David Torres, Dick Walsh, Dioselin Gonzalez, Donnachada Daly, Emilio Gerbino, Esteban Papp, Evan Rosky, Fabio Lignini, Francois Coulombe, Fredrik Nilsson, Gabriela Venturini, Gautam Thakur, Gilbert Davoud, Golriz Fanai, Greg Junker, Guido Zimmermann, Hans Rijpkema, Henna Nandwani, Henry Steinbeck, Hilder Mock, Himanshu Maurya, Jacqueline Chu, James Chun-Chih Wu, Jaryd Snover, Jason Reisig, Jason Schleifer, Jeffrey Mahovsky, Jennifer To, Jose Medina, Josiah Larson, Joyce Dabney, Justin Saunders, Kendal Sagar, Kevin Ochs, Kirk Lansman, Larry Cutler, Liron Topaz, Lou Dellarosa, Luca Prasso, Luis Albertinazzi, Luis Gomez-Larez, Luis Moran, Luke Halliwell, Marc Soriano, Marek Kochout, Mariette Marinus, Mark Hampton, Mark Rubin, Martin Labombarda, Martin Villanueva, Martin Watt, Martin Zenzerovich, Matthew C. Gong, Mauro Cipolatti, Max Kislik, Michael Babcock, Michael Bassler, Michael Hutchinson, Morgwn Mccarty, Nate Reid, Nhi Hua Casey, Nicholas Bertoa, Oliver Bayley, Olivier Staphylas, Othieno Okongo, Pablo Zorrilla, Pablo Zurita, Paul Carmen DiLorenzo, Pete Wilkins, Pia Miniati, Pratik Mankawde, Rajesh Pk, Ramiro Del Corro, Rani Naamani, Ravindra Kumar, Rex Grignon, Richard F. Rubio, Riva Chang, Robert Helms, Ron Griswold, Ryan Casey, Sahil Ramani, Sandy Kao, Satheesh Subramanian, Seath Ahrens, Shantanu Mukherjee, Simon Otto, Stephen Candell, Stuart Bryson, Suelika Chial, Ted Forgrave, Thanh Giang, Tim Ingersoll, Tom Molet, Tristan Malbran, Vishwa Ranjan, Vitaliy Zavesov, Warren Riley, William J. Ballew, Yagnanarayanan Kalyanaraman, Zach Carter, and executive sponsors Lincoln Wallen, Jill Hopper and Jeff Wike.
Dan Sarto is Publisher and Editor-in-Chief of Animation World Network.