In part 2 of our DNEG ‘Togo’ breakdown we explore how the CG dogs were animated, including how dog FACS and even dog motion capture were explored.
Previously, we looked at how DNEG built the CG dogs required for their Togo visual effects work. Now we dive into the many animation challenges the studio faced on the show. Moreover, these challenges included close-up hero dog animation. As well as dealing with a pack of running, slipping, and sliding sled-dogs – often harnessed together – that had to interact convincingly. Plus, the CG animation needed to match with the extensive live-action dog footage filmed in Alberta.
In getting to final dog animation, the production also explored dog motion capture. While DNEG invested time in researching dog FACS. Yes, there’s Facial Action Coding System research relating to dog faces. DNEG lead animators Arna Diego and Leo Bonisolli, who worked under DNEG global animation supervisor Aaron Gilman, break it all down.
Dog mocap R&D
Arna Diego (lead animator, DNEG): Early into production Aaron had arranged for a motion capture session at Animatrix studios using trained dogs. It was a great experience seeing these dogs in action and working with the trainers to capture various behaviors.
As VFX animators we’re used to seeing people shooting mocap, and most of us have suited up ourselves. But it was something special seeing the same process set up with dogs. There are technical challenges in how the suit is fitted and dealing with tracking markers on fur, and generally, it’s a completely different experience working with animals as ‘actors’ trying to capture the right performance.
Below, some dog motion capture from Animatrik (not related to Togo).
It's a wolf eat dog world out there…and in here. More Animatrik studio action 🐶 #mocap #animals pic.twitter.com/WmTdnUqAXY
— Animatrik (@Animatrik) August 28, 2017
It was a truly memorable workday and a cherished experience. We didn’t end up using the data, but it was a good learning experience for any future work.
Leo Bonisolli (lead animator, DNEG): The mocap was more of an experiment to see how well that was going to serve us and how fast they could get there. In the end, it ended up being such a performance-driven show. So doing dog performance with mocap was not really going to be a good payoff. Because you would have needed to modify it so much to get the right level of danger in some of the scenes.
More R&D: dog FACS
Diego: I got that job early on to become very knowledgeable about all the muscles in a dog’s face and how our face rig would work. It turns out there is research study called ‘dogFACS’. So there is human FACS, of course, but the same thing exists for dogs. The researchers have gone in and categorized all the dog face shapes. So we did face shapes based on dogsFACS and we’re continuing to use that now.
You have shapes such as ‘lip corner puller’ and ‘nosewrinkler’. A growl for example is a combination of ‘upperlip raiser’, ‘nose wrinkler’ and ‘lip pucker’ among other shapes. We based our facial control scheme on these shapes enabling the animator to activate and offset the individual shape creating detailed and believable animation.
Our rig for the dog face
It was the result of a really close working relationship between animation and build and rigging, probably one of the tightest that I’ve seen on a show at DNEG. Myself and the modeler, we sat together and studied all the reference and studied the dogsFACS and built these shapes in tandem with each other, giving each other feedback on it. We were also involved with CFX later on.
Mischa Kolbe was the rigging lead who, along with other rigging artists, redeveloped our entire quadruped rig to support realistic performance to a higher standard. This included a new front leg module that has a ‘limb lock’ functionality enabling the animators to mimic the locking of front legs most animals do when carrying their weight over the leg. We also reconstructed our spine setup for improved realism and functionality for the animators. Generally, there was a huge effort from rigging to develop a new and improved standard of quad rig in close collaboration with anim, always working from studied reference.
Snarl-time
Diego: Where the dogsFACS came in most useful was for that close-up shot of Ilsa snarling. On set they had put this prosthetic around her nose and it made her do this fake growl. But her eyes were happy and she was wagging her tail! So we worked very closely on doing that growl and snarl with a CG dog head.
For that we particularly added a lot of small things like little pullers next to the nose to show micro-movements in the muscles. One of the challenges is that FACS shapes and blend shapes are very linear, but a dog’s muzzle is actually really mobile. It does more than just slide around, so we had to ensure it had those proper micro-movements.
Bonisolli: The first time that they showed the anim tests for the snarl scene and ran it side by side with the live-action – that was a really cool moment. 90% of people watching it couldn’t tell which one was which – the fur was amazing!
Slip and sliding dogs: where to begin
Bonisolli: Aaron Gilman got us started by looking very early on at reference of anything that we could find online and of the original plates for all the slips and falls of the dogs, anything that would help sell the feeling of not really being stable on the ice.
As soon as we had a little bit of a library, we started working those out in animation clips. Those clips were roughed in into the shots using the Time Editor in Maya. We could place these as building blocks for each of these slips and falls to establish a preference for each dog throughout all the shots. These helped us a lot in terms of narrowing down what the dog should do for a particular shot. If we considered the amount of detail and animation quality, it could be considered as a pretty advanced kind of blocking.
Diego: We have probably seen every slipping dog there is to see on YouTube! We had to mix with live plate dogs, so we couldn’t make this stuff up. What we found, in general, was that the legs would slide out, and you get that straight leg, it slides out and then their body falls down and then they try to catch themselves up again. We’d start with a run cycle, maybe match to a real clip, mix clips together – it was a bit of a jigsaw.
The art of animating multiple dogs
Bonisolli: When they’re running the sled, there’s 10 dogs, plus Togo. We knew from the very start that the way that a dog is behaving in a team composition is not the same as a dog working on his own. So we developed a rig here at DNEG that helped us establish how far out we could push each dog in the team composition. A part of the rig would turn red if the dog would be pulled too far. It wasn’t a super-accurate system, but it would put us in the ballpark of what we could and could not do in team composition.
And then that made it much more reasonable for CFX to pick up our anim without having to animate ropes that are stretching 20 times the length that they would need to be because we were already that much more close to the real length of what the rope would be able to do.
We incorporated some layout development time into the animation process where we cached numerous different cycles and different speeds for the gaits of the dogs, and they got fed into the rig as something Layout could just switch between, rather than loading the rig and having to animate the dogs running around.
The pre-production work we put in turned out to be a fairly convenient setup for the Layout Department as that allowed them to animate the sled at a speed that made sense based on the speed of the chosen dog cycles, while also retaining a good level of realism as to how fast the team could run.
Stay tuned for part 3 of our Togo coverage. It details the expansive environment VFX that DNEG crafted for the film.
Images © 2019 Disney Enterprises, Inc. All Rights Reserved.
