At the end, I realized that the most important thing in the course wasn't just the amount of knowledge and information, but especially a good and experienced tutor.Martin Demovic
And my compositing journey continues…
Shot 04 : Night cityscape with CG cars
At first glance, this shot didn’t look complicated, because it was shot by a static camera without any big additional effects. However, it wasn’t as easy as I thought because my main goal would be to invisibly integrate two CG cars.
As for every shot, this, along with creating my first long roto character mask for almost the entire sequence, was very challenging. My additional goal was to replace the sky, add elements such as building windows, interiors and neon logos to make the city more vibrant in the evening time.
All materials such as the CG cars, neon adverts, BG buildings and DMP were provided by the academy. It was up to us to choose the material and the way the shot is gonna look at the end.
In this case, I tried to find references for the later evening sky, city skyline and windows interiors. As for the CG cars, I had the references directly on the plate, so from this point of view, it saved me a lot of time.
city /sky references (source: internet)
As I mentioned, the camera was static so it was pretty easy to create the roto mask for the whole skyline. Things I had to fight a little with were bright edges around the city and the moving traffic cables. For the edges, I used edge extensions, edge blur and local edge grade masks to reduce the brightness. For the traffic cables, I used animated spline shapes.
CG car integration
I started with a classic single channel render passes composition and tried to match it with the look of the surrounding cars. Having many render passes really helped me to modify many different parameters to get the best result. For the separate passes, I had shadows under the cars and contact lights on the road. I animated these two layers separately, as the situation required.
The biggest challenge here was to create the real front lights as it was a crucial animated element to the matched look of the real cars. I tried to follow the video in which Vladimir showed us how to create this effect and I spent many hours to get a similar look by using several glows, glints, god rays and grades. Especially problematic were the edges at the moment of the man passing through them, as it caused him to have a very strong light wrap.
DMP, windows, interiors and neon logos
To get a more vibrant look of the evening, I randomly put neon logos on the building windows and interiors. As it was a static shot, I just easily merged it on the top of the background and played with the intensities. I’ve mentioned that I also had the BG buildings extension, DMP. After many unsuccessful attempts to correctly integrate them, Vladimir advised me not to use them at all, because it could mess up the shot completely. After all, sometimes less is more.
I added last color corrections, grain, glows, distortion and lens dirt. Again, nothing is as easy as it looks in the beginning, so as always, after many versions and Vladimir’s feedbacks, I finally got a nice result.
At the end, it maybe doesn’t seem like I spent tens of hours on this shot to achieve the right look. However, the time I invested in it, was balanced with the amount of knowledge I acquired in the process.
Shot 05 : Golden Compass style – keying
This is a classic shot which you can get in every vfx studio in the world. Green screen separation, also know as keying, is also the basic knowledge of every compositor. Honestly, I can say that this was one of my first green screen shots and, as always, it was a big challenge to get it right. As I had a lot of theoretical knowledge from the course on how to proceed, I used this shot as a test to try out all keying techniques that I know. We had many green screen shots to choose from, but I liked this one the most because the hairs are very hard to key in every situation and I had also intended to replace the pocket watch.
I’m not gonna go through all the techniques which I tried to use for this shot, but I will just describe the final ones which worked for me. As I learned in the academy, keying should consist of the core matte and the edge matte for the details. Of course, this is a very general description where we have to take into consideration the many additional mattes to fix the local areas which always appear.
For the core matte, I used the Primatte key which provides a very rough mask, called also inside or outside garbage mask. This removed most of the holes inside the key or noise on the outside. For the edge matte, I used a combination of IBK key and Keylight which are the best for details.
Vladimir showed us how to use IBK key in a more sophisticated way, called the “stacking method”, which really pushed this key to be the major edge key probably in most of the vfx studios.
As I mentioned, these few techniques did 90% of the work for me. The important rest of the work had to be managed with a combination of other techniques, such as edge erode, edge blur, roto mask, grades etc.
Despill and edge fixes
As I learned, despill is the removing of a green or blue color spill from the plate which was caused by a green or blue screen environment on the set. In this case, I used the Keylight despill, but as always, there are many techniques which you can use to fix it. Last but not least, I have to mention the edge fix which is always part of the keying. It is due to a despill, that the original color (green or blue) on edges will change to a different one. It is mostly a problem with dark or bright edges on the replaced background, that could be fixed by some edge extensions, edge grading or eroding.
Background replacement and snow/fog elements
At the beginning, I had an idea to move this shot to a winter atmosphere. I tried to find some nice background of the frozen landscape on the style of the Golden Compass. When I found it, I realized that it looks too boring. So, after discussion with Vladimir, I decided to add some snow elements to get a nice winter filling. The course has taught me how to create snow with nuke particles. So I applied my knowledge in this shot. I used the nuke particles for the background and midground snow and real elements from the school library for the foreground detail. For the fog creation, I used the animated noise separately for the background and the foreground.
As every shot should be a challenge in the process of learning, I had one just in front of me. I have received feedback suggesting that, because of the strength of snowing, the main actress should have some snowflakes on her back, shoulders and hair.
For this, I used the smart vectors, which in this case worked almost perfectly. I created the snowflake texture with alpha and I have painted this texture on the character’s body parts. Later, I rendered the smart vector for the whole sequence and used it with the ST map node to deform the texture based on her movement.
As always, this is the time for finishing the shot with color corrections and optical effects.
Shot 06 : 2D to 3D multi photo projection
The last shot is a particular because it’s an animation created from 16 photos. I used a technique called still camera tracking and using nuke camera track, I calculated cameras for each photo separately, instead of calculating a sequence. As a result, it generated cameras for each photo and I achieved a whole scene.
It is similar to photogrammetry, although not that sophisticated. My goal was to create a camera from each photo which will be used for further model creation and projection.
As Vladimir always repeated during the course, projection techniques are one of the most used in composting, so understanding them and knowing how to handle them is very important.
Still camera tracking
As I’ve mentioned, I used a similar technique as the one for camera tracking, but I used stills solving. The calculation is similar, so you just try to find similar tracks on each photo and, by this, create cameras and points. You can use also manual tracks to help automatic solving get a more accurate result.
Creating a model inside of Nuke with Model Builder
For the model creation process, everybody can use different techniques and also point cloud model as a reference. For modelling, I used the Model Builder, solved camera and photos as a source reference. I tried to build the geometry part by part with simple shapes like the cube, card or cylinder. Sometimes, when the real objects were more complicated, I used Edit Geo node to modify the geometry to achieve more complex shapes.
Camera projections and animation
After I finished the 3D model, I used it for projecting part of the photos on it through the correct camera. As some parts of the objects on the photos were overlapping, I had to use more projections for those objects or do the clean plate from one camera and project it back. Overall, I used 44 projections from 11 cameras which were needed for me. When I finished the model and textures from projections, I created the camera with a simple movement. It couldn’t be too big, because all camera projections have a limited camera angle after which the texture of the model looks deformed.
To get more realism into the animated sequence, I used a motion blur and the DOF from the rendered velocity and depth pass in the nuke scanline render. At the end, I added the camera distortion, grain and color correction.
This project was very demanding, but it really helped me to understand camera projections and situations in which I can use them. Before, I couldn’t imagine that it is possible to create an animated sequence by using just a few photos, but thanks to the course, now I practice that almost every day at my work.
This reel project took me almost a year to finish. At the end, I realized that the most important thing in the course wasn’t just the amount of knowledge and information, but especially a good and experienced tutor. Vladimir led me, gave me relevant feedback and experiences which moved me and my work much further as it wouldn’t be possible to finish it by myself.
With this article, I wanted to give a special thanks to Vladimir Valovic, his vfx course Compozitive and the compositing community he is trying to build.
P.S.: In the next few months, he will be releasing his new compositing course, on a new platform, with new (all in English) videos. So all of you will now have the chance to learn compositing and break into the vfx industry. I wish you a lot of faith, patience and a clear goal.
And of course, don’t forget to be COMPozitive.
Thank you, Martin Demovic.
Martin Demovic / Compositor in Pixomondo Stuttgart
Martin’s IMDB: https://www.imdb.com/name/nm9793848/