Back to Business

F3 Visual Effects at FSU - Week 3

Technically week four, I'm back after a week off for a rocking Spring Break in LA.

For week 3, let's talk about Nuke.

I have been spending almost 8 hours a day in Nuke for three straight weeks now.  My total time spent in Nuke comes out to about 27 or 28 hours a week, which means that since this cycle has started I've spent close to 84 hours or so in the program.  Couple that with the 10 hours or so a week I was spending in the program over the 8 weeks before VFX started for classwork/homework and the experience I brought in from our intro course last semester and it starts to add up to a lot of hours in the program in a relatively short amount of time.  I have really learned a lot about the software and am immensely happy with my progress in it.

This past week I have spent the majority of my time working on crowd replication and 3D camera projection.  These are big steps for me.  The crowd replication shots pushed my knowledge of keying and forced me to start looking for creative solutions.  Here are three things I started doing to my keys that made them stronger keys that looked like they belong:

1.  Edge Correction

One problem I was facing was green tinted edges around the edges of my comp.  Getting rid of them by eroding in sometimes wouldn't cut it.  It took me a while to sort out how to pull this off, but I eventually figured out that I could make a separate pipe off the keylight node, erode in the key to get rid of the edge in the alpha, then invert that alpha so that only the edge was present.  Once I had isolated that edge, I could use it as a mask on a color correct node and get rid of the green by grading it out.

2.  Core Mattes

One problem I was running into was the key pulling data out of the green channel across the entire plate, not just the screen.  This wasn't always an issue; it only became an issue on some of the noisier footage that was shot slightly underexposed.  To solve this problem I had to matte an alpha channel into the plate by using a shuffle node, then feed that shuffle down a separate pipe and merge it back into the main pipe after the key was pulled.  Then I just had to make a rough garbage matte in a roto node, animate it quickly to follow the action, and apply it as a mask to the new merge node.  I don't think this is the best or most accurate way to make a core matte considering it requires me to animate some roto mattes.  However, in this particular shot all of the subjects in my plates were standing almost completely still so this method was extremely fast and the results were great, and I do believe the best methods are the ones that provide the best results the fastest on a shot by shot basis.

3.  Keymixing

The ability to quickly and easily key mix in Nuke is one thing that makes Nuke a superior keying solution to After Effects.  In After Effects, it is possible to "keymix," but it requires either duplicating your plate as many times as your keymix will require and then masking and keying each plate separately or precomposing your composition after each key.  Both equally obnoxious and frustrating.  In Nuke, it's as simple as pulling another key and then plugging both keys into the keymix node and applying a fast garbage matte roto mask.  It's very fast and is just as easy to mix in the twelfth key as it is the second, as opposed to After Effects which gets harder and harder with each mixed key.  Keymixing was a fast and efficient way for me to isolate color variations across the green screen and take care of them.

In addition to these three keying techniques I used to help sell my crowd replication shot, I also added  in a 3D camera move in an effort to help glue the shot together and draw focus to the center of the frame.  In order to pull this off I had to utilize the technique of 3D projection mapping.  In After Effects I have created 3D moves by separating assets onto planes and pushing them back in z space dozens of times, but I had never attempted projection mapping before.  In AE, projection mapping is an overly convoluted and difficult process that I rarely had a need to even attempt, and when it could have come in handy I just chose to look for an alternative solution.  In hindsight, I really should have put the time into learning how to projection map in AE, but in Nuke it was an amazingly logical and fairly straightforward process.  I roughly modeled the scene out in 3D space using planes that I oriented to match the space (ground plane, house, people, etc) and then projected my comp onto the 3D scene using the Project3D node in Nuke.  The more accurate your 3d geometry, the more correctly the projection fits and the more accurate your camera move becomes.  But this scene was so simple just using rough planar geometry was enough to get by.  Again, the fastest way to achieve the result is the best.

I have learned a lot about Nuke over the past few weeks by being forced to think through all of these issues and come up with creative solutions.  I am really excited to see where I go from here with Nuke, and if I continue to learn and progress at this current rate I truly believe that by the time thesis visual effects roll around I will be at the level of total photo real compositing.  I just have to work on my 3D rendering first :)