Producers and recruiters? This is totally what you’re looking for.
This shot was done as part of a short film I shot called “When I Was Young and in my Prime” directed by a friend named Jeremiah Munce. The shot is the POV of a woman walking through the park on a fall day. During photography the presence of park garbage cans and signs was not thought to be problematic. But in editorial I became aware that the shot would be more in keeping with the film if these elements were removed. This exact technique was used extensively in “District 9″ to clean human actors out of shots before the CG Aliens were dropped in. Here’s how it works.
I start by using a 3D camera tracker. Nuke has a great one, but in this case I used SynthEyes. Once I have a 3D camera move in Nuke I go about orienting the camera so that the ground survey points are on the 3D space ground plane. I then use those survey points which are close to my ugly items as a guide to where I place my “cards” in the 3D space. If there is not a survey point near those items, I recommend using the Points to 3D note to get an exact lock on those positions. This will eliminate any drift in the paint out elements.
Once I have my geometry laid out, I put my timeline cursor at the middle frame of the shot. I then duplicate the 3D camera and then wipe out the animation of that camera accept for that frame, fixing it in place on that frame in the middle of the shot. This becomes the projection camera.
My next step is to render off that single frame, or reference it alone. This becomes the cleanup frame from which the whole process depends. I take this frame and clean it up with any paint tool I desire. I can use Photoshop, or Nuke’s own paint tools.
Once this clean frame is ready, I feed it onto the various cards using 3D camera projection, referencing the stationary projection camera. That whole scene is then photographed by the original moving camera, and the result of that 3D scan line render output is comped overtop of the original plate. At this stage it becomes apparent why I chose to work with a middle frame as the key clean out frame. If you use the zero frame, the shot works perfectly at the beginning and distortion gradually gets worse and worse. If you choose a middle frame, the distortion is barely noticeable at the beginning, disappears totally in the middle of the shot and then only slightly reappears towards the end. We are “splitting the difference” and remembering that VFX is slight of hand trickery but can work totally if there is enough movement and distraction to keep the unsuspecting viewer from looking too closely at the pixels.
Once the whole 3D projection gag is working, we are still left with some problems. Namely that the cards are sometimes blocking forground items that we don’t want blocked. We have done an “occlusion paint out” of a background plane, like what we often have to do in 3D conversion. 2d track the elements to be restored in the original plate shot and use that data to drive roto shapes, then comp those element overtop of the original plate and 3D card output.
The hard work is done now, all that’s left is to perform a colour correction and any dazzle elements that might improve the shot even more. In this case I added a sun flare and falling leaves in 3D space. A bit of motion blur in the scanline render node and the illusion works because it also is being photographed by the 3D match moved camera.
In supervising VFX we are often tempted to only use smooth camera moves because they are supposedly easier to deal with. This can be true, but in the case of this shot, the shaky handheld nature of the plate shot makes this trick harder to track and root, but ultimately adds to it’s seamlessness because there is a lot of action and distraction. If the shot was smoother, the viewer is more likely to notice any cards or 3D paint elements that have been inserted. The shakes and the convincing way we are able to use perfect match moving and motion blur in the scalene render better hide andy artificial elements. This is also why you often see camera shake added to a comp once it’s finished, to keep the viewer from locking their eyes on an any artificial pixels.
Nuke is by far the most powerful Compositing tool in existence. This shot proves the point because this entire process is possible to achieve in one package since you can count on NukeX’s camera tracker and it’s paint tools to do this work. In the past it may have taken a trip through up to 4 separate software tools to complete this. Nuke is the most artist and pipeline friendly solution if you asked me and I actually enjoy using it in my spare time.
The clip above is a side by side comparison of a commercial I worked on a few years ago and the previz we created prior. The director is my friend Steve Mottershead who is a rep’d at Soft Citizen and owns an amazing post vfx company in NY called Artjail. We had already worked together on dozens of fx projects and when he won this job, he called and asked if I could help with the planning. He needed to pre-shoot and edit the spot for client approval so that there would be no need to figure out the pacing and blocking during the shoot and there would be no wasted setups. I realized that I needed to recreate the location with “hyperacuracy” so that we could test out how the technocrane would need to be positioned and operated for each shot. This would allow him to know exactly how to answer questions like “where do you want this technocrane?” and “How fast do you want this move” with real assurance. It would also allow him to start the shoot with an approved previz to show, and everyone would see this and know exactly how to best spend their time through the production. No need to decorate anything that would not end up in the spot, no need to exploor how the camera would move at 3am with dozens of frozen tired crew looking on. This plan would provide perfect “togetherness.”
I started by driving up to the location in Markham Ontario, a new development made up entirely of relocated heritage homes. I took a few hours with my Bosch laser tape to get precise scaling of each and every feature of this house. I had my laptop there and whipped up a model right there. The house is not recreated in perfect detail, but the scaling is exactly accurate. My next step was to get the dimensions of the technocrane which would be used in this shoot. This is available on the Panavision website. I rigged the cg camera onto the technocrane with inverse kinematics, so that I could move the camera freely, but I could see clearly in a perspective view if I was driving the crane beyond any of it’s limits. We could also use the perspective views to judge if we were moving the crane faster than an jib op could actually move it. Also we would show these perspective views to the AD and grips to show them exactly where to park the base of the crane for each setup. When I had all this modeling and rigging done, I did a 2 hour session with Steve where we just shot all the shots on the boards in several different ways, always watching that the crane base was parked in a workable place. Any shot idea that had merit was attempted. In that session there is no restriction on time or number of setups we could try. Each time we were happy with a shot idea, I rendered a hardware preview to disk, and we moved on. At the end of this session Steve took the dozens of renders and spent a few days making a pre-cut for client approval. He found the setups which most efficiently told the story in the boards, and found a pacing which worked for him and client.
One may ask, why make these previz shots so crappy and plain looking? Why not embellish them more? I certainly can do better CG work than this but it’s important to keep the big picture in mind. None of this will be seen publicly but more importantly, one should leave much to be desired in the look of the previz shot so that all the artists and designers have room to explore and express their own ideas. What we are concerned with in previz is exact blocking of camera moves with extreme dimensional accuracy and focal lengths so that set dressers and assistant directors know exactly where to put everything with total certainty. Besides, I am charging by the hour so there is just no need for extras that will just waste time. What these images lack in visual quality is made up for by the fact that they are accurate to a centimetre.
Lastly, it is extremely important in commercials to get the pacing right because we only have 30 seconds to tell our story and not a single frame more. When shooting a 30 second spot and referencing a previz the camera crew can be instructed to cary out the blocking in an exact number of seconds. No need to guess at it, it’s already figured out. I have prevized many commercials, but this one uncannily proves the pacing point. Because they were tight for time, very little was shot outside of this plan. Once something is approved by client, a production is extremely averse to revising and breaking that plan. It causes confusion and potential conflict over wasted time and dollars. Having approval on anything is a complete blessing because it allows you to move directly towards completion and there is no need to experiment or invent any further.
I strongly recommend this type of scale accurate previz to any commercial production which is using a technocrane, car mounted jib or motion control because these systems can be extreme time wasters if there is not a solid and approved plan on how they are to be used. Getting pre-approval from Client really expedites things and it’s always best to get things right on set rather than repositioning and doctoring later in post. This process can cost a few thousand dollars, but in the end that’s small potatoes compared to crew overtime or having to do an unplanned shoot day or added days fixing in post. Also, directing a large crew can be stressful, and they are looking to a director for answers constantly. If a solid visual plan is in place, crew can get great answers quickly and feel confident in their leadership.