The art of The invisible effects shot
Boundary Visual Effects (boundaryvfx.com) co-founder Nick Lambert reveals the secrets behind invisible object removal and additions
01 PREPARING THE FOOTAGE
Our story requires a building with two floors. The director finds the car parked on the left of the shot distracting, and the background building too busy with detail. As always, there are a number of ways to accomplish the required results, but we’ll use a 3D approach in Nukex. The first step is to de-noise the footage and take out the lens distortion. If you have a lens checkerboard you can un-distort the footage easily. If not, then a manual method can be used. Nukex’s lens node has a facility to allow you to draw on straight lines and use this to equate the lens perimeters.
In order to get a clean 3D track we will need to mask the car so as to tell the tracker to ignore this area when looking for points to track. A basic garbage matte is all that is necessary here. Roto the car and its shadow throughout the shot. This does not need to be exact, it just needs to cover the offending areas of alternative motion. With the original footage and our garbage mask we can now do our 3D track. This will still require some manual input, such as deleting erroneous tracks and orientating the scene.
At this point, if you have the film back and lens data you can input this to aid the solve, if not, the node will try and estimate these. After running an auto track it’s good practice to delete any bad tracks before solving. Although a solve can be made with relatively few tracks, they need to show parallax to work. Once solved we need to tell our 3D world which way is up. This is done by manually selecting solved points that are on the ground and setting these as our ground plane within the virtual world.
04 SOLVING THE CAMERA
Now we can create a camera, point set and scene for modelling from the orientated solve. Using the point cloud as reference, add a Modelbuilder node to the scene and use a cube primitive to position and scale into the background building. Repeat for the foreground building extension and use a card for the ground plane. We are now set up to start painting and compositing. As you have seen, even a quick and simple-looking shot can require the use of many disciplines, but in a smaller facility this can all be done by one artist in a single application (in this case, in Nukex).
05 FINAL COMPOSITING
Using a combination of frame holds and projectors it’s now possible to paint out the offending car left of frame and remove the busy background on the building. The set extension uses the same setup, but with more work to build the texture. Combining these with the original footage is done by using the Scanlinerender node to output a 2D version of the scene we can merge with it. Now we distort the output of the renderer to match the original footage and the grain. Using the input footage pre de-grain we can now merge the CGI with it seamlessly.