3D scanning a dog is not easy

Part 1 of our behind the scenes look at DNEG’s VFX for ‘Togo’.

In a special 3-part series, befores & afters is exploring the visual effects of Ericson Core’s Disney+ film, Togo. The movie tells the story of musher Leonhard Seppala (Willem Dafoe). His dog Togo, and the true-story 1925 dog-sled serum run to Nome in Alaska.

Here we dive into the build process for DNEG’s CG dogs. Which had to match to the many live-action ones photographed on location in Alberta, Canada. The process included hard-to-get 3D scans of real dogs. Exhaustive reference gathering, a re-jig of DNEG’s fur tools, establishing a muscle and skeleton build and replicating what a snarling dog looks like. Sometimes the dogs were fully digital and sometimes they involved CG head replacements.

Production visual effects supervisor Raymond Chen, from DNEG, and DNEG digital effects supervisor Russell Bowen, take us through the build. Part 2 explores the animation side of the dogs, and part 3 jumps into the extensive environments DNEG created for the film.

First, a test

Raymond Chen (production VFX supervisor): DNEG had not done CG dogs before, but we had done wolves, which is what we did a test with. We also knew we’d need to update our fur tools. What we had before was workable, but when you need to multiply it by 11 dogs or more, it needed to be much more scalable.

Russell Bowen (DFX supervisor, DNEG): We had done Rocket for Avengers: Endgame, and he had eight hundred thousand hairs. But these dogs needed into the millions, the tens of millions. So it was just a different ball game altogether. Our grooming tool is called Furball. We made a number of developments so that our groom TDs could optimize it to get a huge number of hairs on a single dog and deal with having 11 dogs in one shot. It was a big deal to be able to get to that really realistic level while also simulating the dogs with snow, ice, water, and then rendering all of that.

Dog scans: a different kind of challenge

Chen: During the shoot, we had about 30 or 40 dogs scanned by Clear Angle, which was pretty difficult.

Bowen: It was done with photogrammetry in a photo booth.

Chen: Just think about all those cameras and all those shutters and flashbulbs…woah. So, we had to make sure that when they came into the booth, they were comfortable. We’d let them sniff around, walk off and come back in. When it was actually going off you couldn’t let the other dogs see what was happening.

Bowen: We did, though, let them hear the firing mechanism from a distance. That was part of the conditioning before they actually brought them in. But once you brought them in you basically had one chance because they would not come back again.

How it worked was that the trainers – who were incredible – would get the dogs to stand in the pose we needed, as neutral as possible. You don’t want them looking down or up or backward or crouched down. You want them standing as straight as possible.

Chen: Oh man.

Bowen: I just remember standing in that booth going, ‘We’re running out of time…’

Chen: And actually we had to do it twice – once for the summer coat, once for a winter coat. By the second time we did it, we had a condensed number of dogs.

Bowen: And then I had a separate booth going to the Clear Angle booth that I used for getting as close to the dogs. As I could for macro-level photography. We also set up animation reference cameras there to try and capture little nuances and characteristics of the dogs outside of actual plate photography. Which hadn’t happened at this point? We needed it to get going with building rigs and assets.

Dog scans Images

A large volume of the dog sledding shots began with live-action plates.

Here, DNEG retained the dogs but ultimately replaced most of the environment, and added extra atmospherics.

From data to dogs

Bowen: All this data gave us the feet, the legs, the head and a rough volume of the body, but not fur. We knew what was going to become tough to analyze was muscle mass and volume preservation – the stuff underneath all the fur. The way we started on that was actually in that booth again off to the side. I had a little measuring tape and I was literally measuring around the neck, on the back, on the tail, and trying to see how long the fur was at any given point on these dogs.

Chen: I’ve found, in builds like these, a lot of it is based on more idealized versions of dogs, things in anatomy textbooks. It’s not necessarily data-driven. So here we did look at all that, but then we had specific things from the dogs used during filming.

Bowen: What we ended up building was a bespoke muscle system and skeletal system for the dogs. Then we also had muscle and fat systems in Ziva Dynamics. We did skin in nCloth in Maya. And then the fur was groomed in Furball and follicle dynamics did with vellum in Houdini.

The fur was interesting because at some point we realized that a lot of the movement in the fur was actually coming from the fat movement, not the fur itself. The fur doesn’t really flop around. It’s actually quite stiff. It’s so dense that it doesn’t move so much, but ‘separates’ and you get this particular kind of clumping. It’s the fat and the muscle moving around underneath that creates the jiggle and the wobble. Then you add a harness that’s attached to the dogs pressing and pulling on the fur, and we needed to of course sim that as well.

Dog FACS

A snarl rig was employed in attempt to acquire some of the action from the dog.

DNEG replaced the head with a digital version.

The secrets of a (digital) snarl rig

Chen: Togo was our hero dog, but I think some of the best CG work we did was for Ilsa, who had to snarl at Togo. When we were filming, the trainers had worked up a snarl rig. It’s basically a rubber band and prosthetic that holds the mouth open. Usually it would need to wrap around the snout and then you remove the band in post.

The problem, of course, is that dogs don’t really like it. The dog playing Ilsa that they used was good natured enough to actually tolerate it, but was also basically the happiest dog on the planet! So she didn’t look like she was snarling at all. Her eyes were up and her tail was wagging – she did not look like an angry dog!

So I suggested to Ericson this would probably be a good candidate for taking it over and doing a CG dog head on top of the body. We actually did a little proof of concept for him first to show how it could be a little more fierce and angry. That was both a challenge and also a great project for the riggers, modelers and animators to get a handle early on on subtle facial animation.

Bowen: And in the end, Ilsa’s head was actually more detailed and further progressed than Togo’s because we were almost more than full frame on her.

Chen: What I loved about those final shots is that it was acting, basically. Ilsa had to show that she was angry and annoyed and then as Togo starts to lick her to settle her down, there was another level of performance there. An amazing bit of work by our team.