Director Lars von Trier shot Dogville with the Sony CineAlta HDCAM on a large, nearly empty soundstage dressed only with a few set fragments and white outlines on the floor indicating the layout of the film’s tiny Rocky Mountain town. That minimalist approach wouldn’t seem to demand special effects, but the script called for overhead shots – and the set was far too big to be taken in by a ceiling-mounted camera. The production built a ceiling rig full of DV cameras to help Danish effects house Gearless get an impossible angle on Dogville. VFX Supervisor Peter Hjorth explains.

Using the camera rig: We used 13 Sony PD100 cameras at a time, and created about 180 layers [for the most complicated shots]. We moved the rig around 12 times to cover the entire floor, three times down and four times across. The cameras are set to manual, and we had relays built into them so we could remotely start and stop them. I have remote monitoring and a scope, and that’s it. Originally, Lars wanted to have the ceiling full of cameras, with maybe 150 cameras there at the same time. We actually looked into putting photosensors in the ceiling so the whole room would be one camera- but we couldn’t finish it in time.

Camera tests from Dancer in the Dark: The cameras we used for the effects work in Dogville were left over from Dancer in the Dark, when we had 100 of them- strange things happen when you work for a company that decides to buy 100 DV cameras. We had a wall of 100 cameras for Dancer in the Dark, and we did a test composite, putting all those pictures from that wall together. Lars decided to just cut between the cameras, and I was begging him to let me put together some of it in post so we could see what it looks like. So one night, we captured a frame from each camera and put them together very roughly. Out came this strange, distorted perspective.

Tracking motion from zone to zone: Even though the perspective is destroyed, it still looks photorealistic. But Lars wanted people moving around inside that, from camera to camera. I looked into morphing the shots together, and that was just out of the question, both in technical and budget terms. So for the last few days of shooting, we actually painted the floor green and shot everybody who had to move between zones- or between rig positions- and then I tracked them into the large shot in Combustion. We used a Techno Crane, but it wasn’t motion-controlled in any way. It was approximate. If the crane operators make a little mistake and lag behind the action, it’s fine because it will just change the perspective in a sort of random way.

Perspective tricks: If you look at elements in the corners of the shots, you will see things like a chair that has its legs pointing in different directions. The perspective in the shots- it’s just so wrong, most of it. But because it has been captured with a photographic camera and you can put it together without too many seams, something strange happens and your brain says, "Oh, this is a photograph. I can trust this." And all the perspective errors are sort of cleared out. I think you can do a lot of stuff as soon as you throw away perspective. If people pick this up and start roaming free in a world where you can just put [multiple-camera] compositions together- combined with desktop tools that let you put images together when you’re on the set- you can do very interesting things. Of course you can do some very ugly things as well, but that’s the nature of experimenting.

Working in HD/25p: Everything was done in HD, so all my previews were HD, and we shot some previews to 35mm as well. I showed Lars everything in HD. Everything was 25p, even the preview renders. All the I/O and viewing was from the Accom WSD/HD. For Manderlay, [ von Trier’s next film], it will be all Cineboxx review and I/O. Then I did a downconversion for the Avid so the editor could see how the stuff cut in and out. It goes to film one frame at a time, and the sound is retimed. The whole film is done at 25, which is the European style of HD production. It makes everything so easy, because all your monitors and other gear just stays at 50 Hz.

When a bed’s not a bed: When [star Nicole Kidman’s character] Grace is in the fetal position and the camera is rotating, we actually shot a little bit outside the set to get the move that Lars wanted. We did a matte painting of part of the bed and part of the floor. We were asking the matte painter to please do a perspective like 17 different eye points in the same image: "Please mess this up." That was a little uncomfortable for him, but when Lars came in and explained, it was a lot of fun doing it. Everybody can see it’s the bed from the set, but if you sat down and wanted to construct a 3D model of that object that you see in the film, it would be very strange, sticking out in different directions. I hope that, for the next project, we’ll play a little bit more with the options that this system gives us.

Working with Lars Von Trier: In one of the first meetings, Lars said, "Let’s just try and shoot a lot of images from above, and you just try to put them together. And if it looks wrong, please show it to me, because maybe I’ll like it." In general, that’s the beauty of working with Lars. You feel very safe, because even if you do something that, from your own standpoint, is not that big a piece of craftsmanship, he will see something in it that he likes. He will be the filter, which is the mark of a good director. I’ll be doing the VFX shots for Manderlay in early May. From the technical point of view, there are definitely some aspects that are the same- and, of course, there’s something new. Lars always goes forward. He never goes back to anything.

Inside the Truck: The truck leaves Dogville and we zoom to a close shot of the tarpaulin that becomes semitransparent and reveals Grace [ Kidman ] in the back of the truck. Then there’s a whole scene in the truck with Ben, the driver, all with superimposed semi-transparent tarpaulin. And finally the pullback as the cover is removed after the truck stops again. Since the people in the last scene are touching the cover, it had to be shot as a lead-in to this scene. This involved having Ms. Kidman under the tarp long enough to cover the preceding scenes, and the truck was shaken by the ADs to give the impression of motion. So the tarp in the overhead image of the truck leaving is actually the one they pull away seven minutes later in the film. This took a bit of time to figure out. The whole scene was rendered with soft masks and keys for the tarpaulin texture in Combustion. In all, you see between three and 50 layers at a time, and the opacity of the mask changes during the scene.