How Aardman Took Flushed Away into CG

Aardman and DreamWorks built Flushed Away in CG, combining stylised water simulation, custom splash tools and clay-like character rigs.
Animated sewer boat chase scene with claymation characters, brick tunnel background, and splashing green water.

Archive notice: This article was originally published on 12 August 2020. It has been translated from German where applicable and lightly edited for clarity while preserving its historical context.

For the first time, Aardman put the modelling clay aside for “Flutsch und weg” and created a full CG film together with Dreamworks. The reason for this were numerous shots with water. Read here how the studios tamed the wet element and gave the actors the classic “claymation” look. This article originally appeared in Digital Production 01 : 07.

Roddy St. James (voice: Hugh Jackman; German: Ralf Bauer) leads an exquisite lifestyle as a pet mouse in a better neighbourhood of London, until one day the underclass enters his life in the form of the equally proletarian and cunning sewer rat Sid (Shane Richie/ Olaf Reichmann). When Roddy tries to get rid of Sid, he ends up in the London sewers instead. Here he meets rats, frogs and snails who live in a weird copy of the British capital – with Tower Bridge and red double-deckers made of rubbish.

This is how “Flushed away”, a CG animation film by Dreamworks Animation and Aardman Animation, begins. It is not the first collaboration between the two studios, but the first time that Aardman – famous for its stop-motion films such as “Wallace & Gromit” and “Chicken Run” – has worked with Dreamworks to create an animated film entirely using 3D tools.

Water tips the scales in favour of 3D

The reason for this decision was water: “Initially, we thought we were making a stop-motion film with CG effects,” recalls visual effects supervisor Wendy Rogers. “But there really is a lot of water and impressive effects. We had water mist, water splashes, even a water purification plant.”

The water flows through the underground metropolis and cascades down waterfalls. The film’s female heroine, the rat Rita (Kate Winslet/Jessica Schwarz), travels through the canals in a kind of tugboat. Like all the buildings, the tugboat is made entirely of rubbish.

“It is already possible to composited CG water into stop-frame material,” says Sam Fell from Aardman, who directed the film together with David Bowers. “We’ve already done this with ‘James and the Giant Peach’. But because the film is set in the sewers, there’s so much water that it wouldn’t have made sense in the end.”

To create the digital water, Dreamworks used proprietary fluid simulations as well as new techniques developed specifically for this film for waterfalls, rivers, waves, water slides and splashes. The water surface was geometry on which displacement maps were placed. The real difficulty, however, lay in the water splashes.

Busy city scene with animated people walking, shopping, and gathering near green water and storefronts
Another reason for the use of computer graphics was the mass scenes in the sewers. The sequence shows how the crowd in the mouse city is created

3D animated city scene with numerous identical characters standing on bridges, streets, and near colorful futuristic buildings3D animation scene showing diverse digital characters walking along winding pathways in a virtual environmentAnimated city street scene with diverse cartoon characters walking, talking, and interacting in colorful outfits

Tricky interaction

“One of the biggest problems we faced was the scale,” recalls Head of Effects Yancy Lindquist. Water splashes had to be “rat-sized” when the directors weren’t asking for exaggeration.

Another problem was the design. “Wendy and the directors wanted distinctive shapes,” he says. For small drops of water, the Effect Leads modelled shapes; for larger areas, they worked with the liquid simulator to get water into the desired shape. “We were able to precisely adjust the simulator to achieve certain looks,” says Lindquist. “We set where a splash of water should be, how high it should be and how many particles it should contain.”

The trickiest problem for the R&D department, however, was the interaction between the actors, props and ships on one side and the water on the other. Effects lead Scott Cegielski programmed a system for the water splashes, which he also described in a Siggraph paper and which provided the effects artists with “water splash” tools. With these tools, they were able to generate splashes, control the simulation down to the smallest details and create surfaces from the resulting particles. “We generated splashes depending on how fast an actor was moving,” reports Cegielski, “and we simulated the splashes with our in-house fluid simulation.”

And this is how the system worked: before starting the fluid simulation, a proprietary tool converted the surface of an actor into evenly distributed particles – more particles if more detail was required, and fewer particles for less detail. Each particle carried a velocity vector pointing in the same direction as the motion of the actor at that point.

The system then determined which particles to use for the simulation. It took into account the distance between the particle and the water as well as the criteria set by the artists for the speed required to generate a splash of water. The artists fed the remaining particles, which they had modified to control the direction of the splashes and their speed, into the fluid simulation programme.

They also localised the areas where the actors and water moved quickly across multiple animation frames to improve the simulation at these points. To do this, they tracked the centre of each actor’s bounding box frame by frame and derived a vector force from the movement in each case. They then inverted the tracking data. In this way, they removed the overall movement of the actor, but retained smaller movements, such as those of the arms and legs.

In order to achieve the impression of large-scale movement in the simulation, the effect artists used the previously calculated vector force: if the main movement took place along the positive Z-axis, they applied the vector force to the simulation in the opposite direction, i.e. along the negative Z-axis. The result was that the water splashes moved along the pre-stabilised movement of each actor.

The artists controlled the shape of the splashes with a noise multiplier for the amplitude. They also changed parameters by painting texture maps, for example to create more splashes with an actor’s hand than with other body parts.

Model house constructed from various recycled materials including rusted metal, enamel pan, shopping trolley, plastic, wood, and tarp.
All the buildings in the sewerage system are made of bulky waste and rubbish. The designers gave precise specifications for all materials (above). The overview sketch (right) reveals the exact components – from school desks to spoons and forks

Detailed labeled diagram of whimsical house built from recycled materials like shopping carts, oil barrels, plywood, and drawers.

Whimsical illustration of Big Ben clock tower beside quirky, stacked buildings with oversized tire and antennas
The film version of Big Ben was built from a washing machine, a corkscrew and a set of golf clubs, among other things

Splashes with a system

The size of the actors was based on the size of Aardmann’s plasticine dolls for stop-frame animation. For this reason, the water splashes were only miniature in size and the team preferred to render them as surfaces rather than particles. For the surface, a tool first converted the particles into an octree-based density grid and then into a polygon file using an algorithm called “Marching Cubes”. The algorithm had the task of creating polygons based on density for each cell of the grid. Surface-relaxing software smoothed the result. Because the topology changed in each frame, the system used a second file containing deformed vertices for each frame’s polygon file to calculate accurate motion blur. The speed determined the amount of deformation. To integrate the surface of the splash into the surface of the water, the vertices, which were also present in the polygon file, were deformed as they approached the surface. The deformation took place in the direction of the surface normal of the splash along a vector that was tangential to the water surface. This created a bulging edge between the splash and the water surface that connected the two surfaces so that they looked as if they merged in the rendering.

Although Cegielski’s tool contained complex algorithms, the splash system could be artistically manipulated. “Scott’s tool made it easy to animate the splashes separately from the actors,” praises Lindquist.

For the water chase, Effects Lead Alex Ongaro organised the integration of the water splashes, which came from the simulation software, into the water surface, which was a deformed surface. This sequence took three people six months to develop and eight months to produce.

“The main problem was that we had the splashes before the surface geometry,” explains Ongaro. “We had to connect the two software systems with the same colour and create a reflection of the splash on the water.”

They solved the problem by matching the particle colours to the water colours and then creating pseudo-reflections in Dreamworks’ in-house compositing software.

“It was a big challenge to create effects that were stylistically exciting,” reveals Wendy Rogers. “The water is painted rather than real, but it’s real enough to look menacing. It’s not entirely dissimilar to the way effects are used in feature films.

3D animated mouse character with brown hair, large nose, expressive eyes, and blue facial animation markers
The mouth animation is crucial to the Aardman style. Clearly recognisable here: the outer and inner band for the 3D animation
3D cartoon character head displaying twelve different facial expressions, including happy, surprised, sad, and confused emotions.
While Aardman used different plasticine mouths for the actors, this time the animators had various 3D mouths to choose from
3D animation interface displaying various expressive toad face presets for character head and eye movement selection.
A special tool provided the animators with prefabricated facial expression elements

A rig for the Aardman style

Actually, stop-frame animation is a kind of real-life feature film, and the directors wanted to bring the famous Aardman style into the 3D world as much as possible. For example, the actors have the same eyebrows as Wallace & Gromit – a kind of “monobrow” – and the same giant mouths.

However, it turned out that this was not so easy for the character riggers to realise. “The directors wanted detail control,” reports Rogers. “They wanted to be able to push in one place on the face and leave an indentation in the eye socket.”

Martin Costello, the technical director responsible for the digital actors, and his team took five months to set up the rig for the mouth and another six months to work with the Aardman animators. “It was one of the most difficult rigs I’ve worked on because it had to be so simple,” he says. “Normally you focus on developing a good backbone and an IK system for the legs. In our system, that was secondary.”

When the animators mould the mouths of their plasticine puppets in a typical Aardman production, they use special tools. Dreamworks also had to supply similar tools for the CG models. As with the plasticine puppets, the modellers created ten different mouth shapes for each of the CG actors. The riggers then provided the animators with numerous controllers around the mouth so that they could actually achieve the numerous moulds.

“In short, there was a narrow band around the lips and a wide band around the cheeks,” explains Costello: “The animators could press the control vertices on the band.”

For the eyebrows, they chose a layered approach. Three sets of joints – left, right and centre – gave the brow an approximate silhouette. Above this, similar to the cheeks, were controllers that could flatten or puff up the front of the brow. “Basically, the animators could model the brows as if they were using clay,” says Costello.

For the body, the riggers created sleek armatures to mimic the stiff look of the Aardman puppets. “Squash and stretch was strictly forbidden at the beginning,” recalls Costello. “But we worked a bit on the arms and legs so that they didn’t stand out so much.”

Between them, the riggers set up eight main actors, ten supporting actors with important roles and 50 different actor types for crowds. To speed things up, they developed a character rigging system.

Automated rigging

The riggers were the first to place the joints. “Assuming you have prefabricated components [body parts], all you have to do is press a button and all these joints appear on the screen,” Costello describes. “You move them into place, click ‘Save’ and they’re all saved in one file.”

Next, the riggers added behaviours to the joints. “If you want an elbow to swell, you have to build that in with the joint positioning,” Costello explains. The subskin – a cylinder of rings that uses a low-resolution Nurbs geometry – moves on the actor in a process that weights the subskin with the behaviour joints. “It’s the same thing again: If a component is already produced, you click a button and it opens a subskin in Maya,” Costello explains. “You can drag the points around, scale them and deform the rings to make an arm swell.” Another click of a button saves all the CVs of the skin position to a file.

The riggers then paint the binding, i.e. they describe how the subskin moves the high-resolution geometry. Martin Costello gives an example: “We were able to paint weights for the shoulder subskin and the upper chest area to give more weight to the chest or shoulders.” In the end, all it took was the push of a button to build the actor with all files and using all states.

“The system worked so cleanly and elegantly and was so consistent that we were able to put together a background actor without having to worry about the details,” Costello states.

3D animation character modeling process showing green frog in purple robe from wireframe to fully rendered stages.
The development of the toad from the simple skeleton (top left), through the controls, lowres, subskin and hi-res to the final rendering (bottom right)
3D animation control panel interface showing character rig hierarchy for body, arms, spine, and selection options
The body controls used by the animators to control the rig

Further development of the Aardman style

Later in the process, the actors developed their own style, a mixture of 3D and 2D animation. About 30 animators worked on the film, and most of them had a 2D background. “They influenced the style,” says animation lead Jason Spencer Galsworthy: “You can clearly see that the actors are based on Aardman characters, but the animation has a slightly more fluid style. We wanted the actors to feel like they were solidly screwed together, but we pushed the style beyond that. We knew it would evolve into something of its own.”

Even though Dreamworks managed to provide the Aardman team with stop-frame actors in a CG world, it’s unlikely that Wallace & Gromit will take the step into a full 3D world in the near future. “We like the fact that stop-frame animation is very direct,” explains co-director Sam Fell. “It’s like a real film set in miniature. If you want to move a shadow, you go and do exactly that. Whereas with a CGI pipeline, if you want someone to change something in an image, you have to write an email. This then leads to a rat’s tail of emails, and in the end you have a meeting.”

Fell, of course, also sees the advantages: “On the other hand, the wonderful thing about the CGI process is that you can work on a shot until you get exactly what you want. You can work on the animation in much more detail and also be much more precise in the art direction.”

Despite their individual advantages and disadvantages for the project participants, the working and design tools – whether CGI or stop-frame animation – remain secondary for Fell. “At the end of the day, the only important thing is that the viewer forgets how it was made after five minutes because they are so engrossed in the story and its actors.”