Alain Bielik meets the vfx wizards behind the new fantasy family film, Nanny McPhee, to try and uncover their magical secrets.
A dysfunctional family, irrepressible children, traumatized nannies until one day, a very special nanny shows up, sets out to perform some magic, and pulls the family back together. Happy end. Remember? 42 years after Mary Poppins enchanted millions of children around the world, Universal Pictures releases Nanny McPhee (Jan. 27), a new take on the pitch. This time around, the nanny (Emma Thompson) doesnt have the good looks of Julie Andrews, nor cant she sing Supercalifragilisticexpialidocious. Yet, magic keeps happening around her, although director Kirk Jones was keen on not having her actually perform obvious magic on screen. Unlike Mary Poppins, Nanny McPhee seems ambiguous about the magical nature of the character, says Andrew Rawling, CG supervisor at Framestore CFC (FCFC). Kirk wanted to have real situations in which the viewer would get the feeling that something was going on. We had a lot of natural events that went a bit different, just a bit. For instance, whenever Nanny McPhee hits the floor with her stick, it creates magical sparks, but we made them look like the contact between the floor and the metal end of the stick could have generated natural sparks This particular animation was created in Houdini, although we could very well have used Maya.
Even though little on-screen magic was required, FCFC still crafted 150 vfx shots for the movie, under Mike McGees supervision. Most of them were invisible effects, beginning with environments. The Brown familys house was created as a full-size set, but the production only built facades for the set of the nearby village. Digital matte painter Kevin Jenkins complemented the partial sets and extended the village with a church, 17 extra houses, gardens, walls and trees. We had helicopter shots in which you could see the whole village, Rawling observes. We first tracked the shots in Boujou and Matchmover, and then modeled simple 3D geometries in Maya. The CG facades were then carefully painted in Photoshop to match and extend the real sets. We used lots of reference photographs of architectures, windows, doorways and all sort of details to create a typical village ambiance. The matte paintings were then projected onto the geometries from the camera point of view, and the various layers combined in Shake, our main compositing package.
Wreaking Havoc in 2D
Several invisible effects shots involved placing the Brown children in outrageous situations, or simply in uncomfortable settings. This included a scene in which the baby girl is found laying inside a cooking pot full of cabbages! To keep the baby as comfortable as possible, we filmed her from above on a soft mat and keyed/rotoscoped her from the background, recounts digital effects supervisor Gavin Toomey. She was suitably smeared with a little practical effects gravy on the day, but all of the cooking pot element was shot separately and composited around her in Shake. Soon enough, the baby becomes the joyful subject of another experiment: the children build a catapult in the kitchen and explore the laws of gravity with crockery, chickens and their baby sister. We actually got the babys father, dressed in a green suit, to hold her up against a greenscreen, and we filmed various takes of the baby slowly moving in an arc motion. We then found the best frames for the action (and expression) and applied a 2D move and natural looking motion blur. The baby was giggling away while we shot her elements, and I think it helped the comedy factor of that sequence. For safety reasons, we shot all the flying crockery as separate elements. We kept the process as simple as possible by using physical effects and compositing that in with a difference matte. By avoiding any need for CG elements, we also avoided the implications of on set interaction, as items shatter against the kitchen set and the objects within it.
The baby girl also necessitated subtle effects work for shots in which she was supposed to speak. Using Inferno, 2D artists selected takes in which the babys mouth made the proper shape for particular sounds. The mouth was then extracted from each selected take and subtly morphed with other mouths shapes to create the illusion that the baby was pronouncing words. A completely different approach was used for a scene in which a donkey starts acting like a human being, sending a kiss, for instance. Since the scene was much more cartoony, FCFC opted for a full muzzle replacement. The donkey was wearing a leather strap across its face, so we used it as the boundary for our CG replacement, Rawling notes. It really helped a lot to hide the transition between the CG muzzle and the real head. The animation was created in Maya and rendered in RenderMan. We had about 10 different mouth shapes, which we blended to animate human expressions. The donkey was actually an animatronic and rod-controlled puppet built by Animated Extras. For the scene in which the animal is dancing, puppeteers animating the body, back legs and the supporting rig and wires were removed from the frame and replaced with a clean background.
Beefing Up the Eeeek Factor
Things turn really magical for the Brown children when McPhee obliges them to swallow a disgusting-looking medicine. This was a difficult series of shots, Rawling explains. In the beginning, we werent quite sure how far we wanted to go with this effect. The plates had been shot with a real liquid pouring out of the bottle and with the children swallowing it. It was a horrible, thick, green liquid I wouldnt drink it myself! So, we didnt want to lose this quality. We thought we could make it behave like it was alive, like a blob wanting to get into the childrens mouth. The problem was that we had a very small image area to work on. It was not big enough to make any subtle animation. We ended up adding big, fat bubbles on the surface, like in a swamp. We first tracked the spoon in Maya and created the swamp effect in Houdini, using the real liquid as a texture map. The liquid pouring out of the bottle was real. Finally, the sound people edited some great sound effects to help sell the shots. It worked really well.
The movie ends on an even more magical note with a highly unusual wedding. Although the event is celebrated in summer, a snowfall turns it into a truly magical moment. The snow transforms not only the landscape, but also the brides dress. FCFC started by turning the green scenery into a winter wonderland. The production used a lot of practical snow to cover the set, but there were still large areas of the background that remained untouched, Rawling says. We first modified the background by digitally grading the image (FCFC graded the entire movie), altering the landscape and the sky separately. Then, we used matte-paintings for the areas that still needed snowy touches. The falling snow was created in Houdinis particle system.
Frosting Up a Bridal Gown
The transformation of the brides dress proved to be the most difficult effect of the movie. The first challenge was to define the nature of the transformation. Was the ice actually becoming the final white bridal gown or was it covering the original blue dress like a cocoon and turning it into the white dress? We had a lot of discussions to determine what was happening. In the end, we decided that the ice would form a cocoon around the blue dress and magically transform it into the second dress. When the bride pulled her veil back, there would be an explosion of frost and crystal, revealing the new dress. The next question was: how were we going to realize it? We considered shooting two identical plates of the actress, one with the blue dress and one with the white dress, but it would have been too hard to line them up. Finally, we decided that we would shoot all the plates with the blue dress, except for the final shot, and work from there. Since there were several elements from the first dress that were not part of the final dress epaulette-like ruffs, for instance we removed them from the costume to shoot the plates and replaced them with CG replicas built in Maya. They were then animated to crumble away in an elegant manner.
3D artists utilized Maya to track the actress movements, using orange markers placed on her costume. Since the fabric of the dress made it too complicated to track precisely, the match was approximate. Using this data, a cloth simulation was run in the Houdini version of Syflex: it created the animation of the skirt, of the end of the sleeves, and of the veil itself the actress had been shot without a veil. Close-fitting enough, the rest of the dress didnt require cloth simulation. Built from reference photographs of the actual costume, the CG white dress was then tracked onto the actress blue dress.
At this point, FCFC started focusing on the transition effect. The first stage involved creating snow swirling around the character and landing on her dress. We did it in reverse, Rawling reveals. Using Houdini, we first attached snow particles to the 3D body and animated them to fly off from the bottom to the top. The animation was then run in reverse to get the effect of the snow landing. It was far easier to do it that way than trying to have CG snow collide with her in a satisfying way. That would have been very tough to control and rig. Working in reverse, we had complete control over where and how and when the snow would land. Then, as each snowflake hit the dress, it melted away, forming an ice surface. For this part of the sequence, we put all our snowflakes back in their start position on the dress and started from there. We generated the wipes in the UV space of the texture map and had the CG flakes spread out and fade. This was actually a recursive compositing technique: we took each snowflake hit and had them slowly spread out of the surface. That gave us a mask of which we used several stages to build different layers of this frosty dress structure. We started off with very faint icy patterns and had them build up and thicken, until we obtained a real thickness on the frost layer of the dress. At the same time, we used Syflex to extend the original dress out at the back to create a train.
The final stage of the effect involved the actress pretending to pull back her veil a CG cloth simulation. Particle animation made all the frost particles shatter and burst off, revealing the real white dress underneath. Throughout the dress sequence, the CG animation was rendered with Mantra.
For Rawling, Nanny McPhee turned out to be a particularly pleasing project to work on. When I saw the finished movie, I didnt study our shots, checking what was right, what could have been better, etc. I just enjoyed it. It was probably the first time that I could watch one of our movies without focusing on my work!
Alain Bielik is the founder and editor of renowned effects magazine S.F.X, published in France since 1991. He also contributes to various French publications and occasionally to Cinéfex. Last year, he organized a major special effects exhibition at the Musée International de la Miniature in Lyon, France.
