CG Laid 'Bear'
By Rob Goodway
March 27, 2015

CG Laid 'Bear'

Paddington is no ordinary bear, nor does he look like the typical teddy bear. Anyone who knows Paddington has a good idea of what he looks like. Floppy hat, duffle coat, button eyes — details you might have variously picked up from Peggy Fortnum’s original illustrations, the BBC’s stop-motion animation, or the rows of wellington-booted cuddly bears that stand in toys shops.

Despite the many looks the anthropomorphised bear has had since his creation by Michael Bond in 1958, Paddington has never had to look real, or interact with real human characters. That is, until Heyday Films and Director Paul King decided on a film adaption of the famous books. And that’s where Framestore (www.framestore.com) entered the picture.

“It was challenging taking on such an iconic character,” says Andy Kind, Framestore’s visual effects supervisor on the film. “There are a lot of looks for Paddington, and everyone has their favorite, so we spent a lot of time on the design before we found something that hit the mark.”

The film tells the story of Paddington, the polite bear from Peru who is invited home by the Brown family after finding the creature in London’s Paddington Station. 

The live action was shot at Elstree Studios and on location in London as well as Costa Rica for the jungle scenes. During the shoot, the VFX team back at Framestore was hard at work perfecting the look of Paddington. “There was a lot of responsibility finding a photorealistic design for a bear that people have so clear in their minds,” says Animation Supervisor Pablo Grillo. 

According to Grillo, the group was keen to bring him into the real world so he would sit into the live action and, they hoped, people would connect with him. “Being anatomically correct meant we needed more detail compared with the simplicity of the original designs, which often had just two dots for eyes,” he explains. “We had to think about things like his wet nose, teeth, and muzzle, but still make sure that what we created carried that simple essence of the original Paddington.”

King briefed the animation team on who Paddington was before they started work, focusing on every nuance of his personality. “We tried to maintain this spirit in the way we crafted the performance. The hope was that rather than filling every moment with movement and unnecessary detail, you could just run a little wind through the fur to keep him alive and really hold some quiet moments,” says Grillo. “In contrast, we have his reactions to some of the more silly situations – to water, to marmalade, and to the various conditions he’s put in. A lot of the humor of Paddington is watching him being put through the ringer.”


Fur Real

Furry creatures aren't new in VFX, of course, but when they are the stars of the show, you know you need to be top of your game. For digital fur, Framestore uses a proprietary hair system, fcHairFilters, which was originally developed for the polar bears in The Golden Compass back in 2007. At its heart, it’s a node-based evaluation framework where nodes (filters) can be connected arbitrarily to form a directed acyclic graph where the hair data flows along the edges and gets manipulated by the nodes. In addition to hair or feather data, the nodes can also receive other data, such as geometry or functions, which change filter parameters on a per-hair basis.

The system has been developed continuously since 2007, and for last year’s crop of hairy movies, Framestore introduced a fourth major version, which substantially sped up the grooming process.

The improvements enable the artists to decouple memory requirements from the actual number of hairs in the groom, and a groom can be computed with a small memory footprint that is independent of the total size of the groom. Other changes have included refinements to sharing memory between filters and improvements to allow multiple artists to work on the same character.

One challenge during the creation of the fur effects on Paddington was the many interactions of fur with other objects that are in contact with Paddington or the other bears (including, in one sequence, Sellotape). To handle all of these shots efficiently, the crew extended their tool set so they could generate data directly inside the fur system and then pass it along to other software packages. 

Typically, image maps are passed into the fur system to control certain aspects of the fur, but now it is also possible to write hair attributes back into a map and then use that map outside the fur system, such as for influencing simulations. Another new feature is the ability to create particles on hair, which can then carry data from its parent hair out of the system. Such particles can serve as locators for instancing geometry (bread crumbs, twigs, leaves), they can serve as starting positions for droplets inside fluid simulations (including the shower in the movie), or they can be used as indicators where a particular part of the hair actually is on a certain frame.

Framestore also expanded its set of tools to allow TDs to do final finishing touches to parts of a groom, as it can always happen that the simulated hair doesn't exactly behave the way it is supposed to and the shot still requires some manual intervention. As a result, a quick way to select the offending hair and some means to change it locally are necessary. Besides simply using the same tools a groomer would use, it is also possible to make the hair visible to Autodesk’s Maya and then apply Maya tools, such as a deformer.

After grooming a furry character, the groom data is brought into fDynamoHair, Framestore’s in-house fur-simulation tool. The studio has substantially updated fDynamoHair over the previous year, devoting considerable time to improving how the artists detect and resolve collisions, and how the simulator takes advantage of multi-threading.

However, with the increased complexity of the Paddington groom, and quantity of simulation controls, the fur-on-fur interactions became a bottleneck. So, for Framestore’s Peruvian friend and family, the crew integrated a new algorithm to mimic the internal repulsion and friction forces of the groom. At each step of the simulation, density and velocity fields are created and stored on a sparse volume. Finally, the friction and repulsion forces for each hair are computed from those volumes that are faster than the naive hair vs. hair collision.

The work is most noticeable where you see furry characters in close contact with each other (as they hug, for example), with digital clothing (especially the famous duffle coat), and in shots involving fast motion.


Water Work

Paddington’s appetite for accidental destruction results in long moments of physical comedy, and some difficult VFX moments. “He gets wet an awful lot – in the rain, in the shower, in the toilet. Those sequences were the most complex tasks in terms of visual effects, and there were a number of them,” says Grillo.

Fur and water are difficult to create individually, so combining them multiplied the challenge. With the elements simulated using different propriety solvers (fLush for water and fDynamo for fur) that had never needed to “talk” to each other before, a whole new workflow was required to allow interaction between the two.

The VFX team began the process by generated a water simulation on top of a hairless Paddington model, thereby creating a wet map, showing in black and white where Paddington comes into contact with water. This was then passed to the creature effects (CFX) team, which used it to drive the fur, making sure that in wet areas, it was made to fall and clump a lot more. As soon as the fur dynamic was done, it was passed back to the effects group to place droplets into the fur and the water dripping from the fur. The look was later perfected in lighting and comp, where the teams added extra droplets in 2D at key moments.

There was a lot of fine-tuning that had to be done, as CFX Supervisor Juan-Luis Sanchez: “The first issue was determining what Paddington looks like wet. We [created the look] as it would technically happen at first, but then we lost some of what Paddington was, especially around his eyes, so we had to go back and re-cover his character. We blended between two or three different versions of his fur, which ranged from dry to soaked, depending on where he was getting wet. The important thing was to hit that balance between realism, the character, and the comedy.”

Of course, all this water results in the classic “animal in a bathtub” moment as Paddington shakes himself dry in two ultra-detailed, slow-motion shots. “We really had fun with it and played up the comedy of his jowls shaking and the water flying off,” says Sanchez. “We added something extra with an fLab (Framestore’s in-house fat solver) simulation, to give the ears that extra level of flap that you see.”

Key to selling the shot was getting a genuine reaction from the kids. “It wasn’t a high-tech solution,” says Composting Supervisor Anthony Smith. “We got a broom handle, about the height of Paddington, with mop heads fixed to it all the way up. We put it in the bath, soaked it, and spun it around. So we had real water hitting real kids and a real reaction. In the comp later, it meant we had to remove this spinning mop-head contraption we had built, but it was the best way to do it.”

A shot of Paddington licking the face of youngster Judy Brown was accomplished in a similar way, with Grillo creeping on set to delicately wipe a paintbrush the size and shape of Paddington’s tongue across her face, leaving the perfect, slobber-like mark. The approach meant we had to deal with the consequences of having to remove the paintbrush, but it avoided intensive facial tracks and simulations of the liquid running down the actor’s face.

After Paddington shakes off much of the water, the Brown children use hair dryers to finish the job, which required careful placement of Paddington to make sure the actors were aiming in the right place, and some tweaks to the groom to make his fur longer so it would show the effect more obviously of the hot air. The end result is pure comedy, as Paddington is turned into a three-foot fur ball. 

“It was just one of those crazy shots where we had to push it a bit further,” says CG supervisor Ben Loch. “We didn’t even go through our normal pipeline for that; we simulated every single hair instead, so that was an extremely heavy sim.”


Dressed for Success

The introduction of Paddington’s famous blue duffle coat was another big VFX challenge, requiring an invisible transition from live-action to CG coat as it is swept around Paddington’s shoulders by the Brown’s eccentric relative Mrs. Bird (Julie Walters).

“We knew this would be a big shot, so watching Julie on set was a bit nail-biting, knowing that just a few seconds of her movement would determine the difficulty level of what had to be done,” says Smith. “We could have had a stand-in to receive the coat, but we decided to use an eyeline stick to prevent obscuring Julie, to prevent lots of reconstruction in post. The eyeline stick gave Julie a good reference for Paddington’s position and height, and fortunately, she did a great job of wrapping the coat over the imaginary shoulders.”

The process of transitioning the live-action coat into a CG version that wraps around Paddington was a real collaboration between departments. “The moment Paddington starts to interact with the coat, it had to move in a way other than just dangling, as it did on set,” adds Smith. “So we tracked some key parts of the coat – the section Julie was holding and the fabric between her hands, along with the front sections of the jacket. We determined we could keep these live action for as long as possible and use them to help reveal the CG coat exterior as they closed around Paddington. All the other parts we would simulate with [Maya] nCloth.”

As Paddington puts his arms through the sleeves, all parts of the coat that would be influenced by that interaction are simulated. The team began the composite with a first pass at this and let the character effects artists know where the transitions – which happen in stages on different parts of the coat – were working, where they could be improved, and where the compositors could assist by re-animating Julie's arms into better positions, for example. 

“We had to have a very productive feedback loop going between comp, CFX, and lighting to get the best results,” says Smith.

“It’s moments like that that I’m most proud of,” says Kind, “where hopefully no one will really notice what we did, but maybe wonder afterward how it was done.”

Rob Goodway is a writer employed at Framestore.


Paddington’s Moving Performance

No one can deny that the main character, Paddington, is cute and cuddly…and digital. But Framestore’s job of didn’t end at modeling and texturing the lovable bear. In addition, the team had to perform the character convincingly alongside real actors. 

To do so, the Framestore animators employed Vicon’s T40 optical camera system to explore the movement of Paddington Bear during the pre-production and pre-visualization stages. The motion-capture system provided the animation team with real-time feedback and enabled them to visualize and develop the interactions between the cast and the computer-generated bear.

“The Vicon cameras enabled us to act out entire scenes in real time during the pre-visualization stages,” says Gary Marshall, motion-capture supervisor at Framestore. “This provided us with the flexibility to guide the creative process and produce key poses and illustrate movement with greater accuracy. Using the system gave the team instant feedback on which scenarios were working well in a shot, and also how the timing and performance of the different mocap performers vastly affected the character of the bear – ultimately advancing the character development of Paddington.”