Some Nuke Scripts/Node Trees

Here are a selection of screen grabs of Node Trees from Nuke:

These examples show the compositing processes undertaken on these particular shots.

Motion Vector | Motion Blur

I have utilised mental ray's pass mv2DToxik to create a motion blur in the shots.

The motion is mostly through camera movement.

The other option for this was to use After Effects CC Force Motion Blur effect as demonstrated by my classmate. However, I want to keep this all within Nuke, so it was important to get the pass to work.

The vector frame

It is a 32-bit (float) vector sequence which you use in Nuke to determine how much the image should blur in each frame by using the colours it provides. Shuffling out the image and setting it to 'forward' and then copying this into the comp in order to provide for the vectorBlur node allows this to work. In this node I set the method to 'forward' to sync this up and the shuffled UV channels to 'motion'.

The vector blur applied

Pulling Keys

I have used IB Keyer for this. It provides a lot more for the hair. I used the luminance match from the background as well as the chroma.......

The shirt area is roto'd and has and EdgeBlur which is brightened to give a soft and light outline....

Overall, pretty happy with this key.

The image will be properly colour corrected using other shots in the film as reference.

I am rendering most of the shots through the farm at the moment, and compositing some other final shots.

Another Ladder

This is another test comp for the Factory scene:

POV of Tyler looking for the exit..

The plan is to get as many of the test comps underway as soon as possible, and capture the feeling for the scene, then to do final renders with some more passes (eg. motion vector). Then the composites should be tweaked and finalised.

As I add more atmosphere to the alleyway shots in order to keep with the feeling of the scene, it is then important to make sure these are as consistent as possible in look. Afterwards, likely requiring dirtying up these CG shots to make them appear as being shot..

Comp-fu:

After getting the bulk of the compositing done, I found that the ladder just didn't stand out enough. This was primarily because of the colour correction washing it out to a blue tone like the rest of the image. In order to fix this I used the depth pass and graded it in order for it to give me a solid white isolation of the ladder. However, the foreground was also white, so I roto'd this area out. Once this re-painting was complete, I used a Keyer node the make the luminance into an alpha and used the Copy (alpha) node to isolate the ladder area of the composite to that point in to re-grade to a redder tone (even with a yellow glare and glow on the highlights - again by using the grade node which is very opaque) and Merge on top again. This brought the ladder to a better colour in the comp, and therefore stand out some more.

Keying and CG Character Comp Test

Keying

I have begun keying the character in some shots.

I began with Keylight and isolated separate areas, such as hair and body, and then merged them back together.

It was getting a little messy, and so I began again with IBK. Despite not having shot the blank green screen I was able to create a very smooth and consistent background using the IBK Colour node.

After tinkering with this process, I found it much more comfortable to work with. I tweaked the darks and lights for some time. Then using background luminance and chroma it gives the colours of the rendered background (or Constant in the meantime).

I also had to switch on 'autolevels' to pinch the whites in more and the 'luminance match' which needed the 'screen range' pulled down.

There were artefacts in the key, so it appeared that the red weight was too low, so this was pulled up to a value higher than 1.

For a better outline, I have premultiplied the key, then unpremultiplied it to get rid of the black outline, eroded the line, then blurred the edge (with a brightening effect.)

Lighting

When I rendered out the roughly keyed footage, I took it into Maya to match up the lighting. It was all very dark, so I created an area light with a warm colour to match the footage as much as possible.

Here is a test render:

After the test composite, this is what I have:

I have used light wrap, a glow and a glint (rotated 90 degrees) which really helps to tie the character into the scene using the lighting here. Of course along with a series of colour corrections and composited elements.

I have chosen to move the camera slightly to the right as the character leans forward.

More Tests

More in synch colour correction

Error at wall fixed by rubble geo (texture and lighting did not do enough)

Very rough comp, but risen 'values' for glowing effect

More feeling of fog and depth needed

Tested motion blur:

Using motion vector pass

Long Shot Test #2

Only master beauty, ambient occlusion and basic zDepth passes were rendered for this.

Power of the comp.

CG Comp Test

I have produced this test, by improving an original shot which looked a little bland.

I textured this with mia material x passes with a light fractal bump, and have lit it with a HDRI combined with directional light and an area light. I have rendered the beauty, specular, ambient occlusion and indirect passes, for example, at 32-bit (float) OpenEXR. This is composited in Nuke to create as nice an image as possible to sell this company.

Working with the depth pass and zBlur node was good practice. Making sure to use this pass and bring it in using the ShuffleCopy node setting the Z to Red and the drop down menu to 'depth'. This allows one to use the focal plane setup to adjust where the depth effects more.

I used the same depth pass to create a light fog by grading the image in the Grade node and merging it. This was tricky, but upping both the multiply and gain together (or exceeding the allowed amount as the effect is similar) allowed to get the right feeling. I also had to invert my depth pass. Instead of using the merge node to combine this, I used another Grade node (which I coloured slightly) and this went into the 'mask' of the node.

Seemless Texture Development

I have been working on the edges and corners of the environment. After speaking to supervisor Dr Jin Zhi, due to the fact that the edges are far away from the camera, one can opt for dealing with the jarring edges through simply texturing. I was going to be doing this synchronising anyway, but it really does work. No need for compositing elements and tracking it anymore!

Here is the latest:

Direct your eyes toward the righthand side of the frame where the edge of the platform meets the wall.

More on this elsewhere, and then I am wanting to perfect the light fog and feeling through lighting and render passes. I think it needs some spotlights, then I'll deal with depth and compositing a 'thick' feeling of fog/atmosphere.

Also, in the alleyway scenes, it was suggested to make some of the fog thicker. Jin says this should keep with the original concept and feeling of the piece where atmosphere and mystery is key. So, going to revitalise this soon. And, make sure these two scene sync up nicely...

Factory/Station Progress

This is what I had yesterday from this angle:

Here is an update from today:

I have used the same composited photographed elements (slightly rotated to fit better) with the new textures, as well as testing some depth and grading. These shots also have a lens distortion and grain applied from one of the lenses we shot the live action on.

For extra details in this shot (as well as some textures), I have used this reference material:

Taking from this reference material adds extra details to the modelling and makes me question the previous shapes and how the structure in fact works in reality.

For example, the extrusions added make the shape along the top of the wall a metal gurder - which then holds up the wooden rafters, to which the windows and window frames sit.

This was not the case previously. And, for me, I don't want it to be made of cuboids in the general area that it should be to look okay - it's my job to make sure these are real objects. I want the feeling of a place which doesn't exist, but obviously one which may.

I look forward to the tweaks to the environments to be made and the textures blocked out. This will be succeeded by adding extra details into textures and hopefully getting newly modeled elements introduced to create more realism through detail. After the final lighting, rendering and camera angles, I will be at my homely helm of compositing different elements together and tweaking the final outcome for the shots.

CG-LiveAction Comp Test

I have tested the idea of compositing live action elements into the CG environment.

Here is the sketch:

This example has partially been textured with the wall on the right being composited from a photograph as well as the roughly roto'd circuit box on the far wall.

I think this is the best option if the tracking process goes well. I see the plan of action being exporting the camera from Maya and bringing that into Nuke for the slight pan to work nicely.

Curved Edges

In the environment I was given, the edges of the room are very hard-edged and in perfect right-angles. In reality, this is obviously very rare, only ever really seen in maths textbooks.

In my previous projects, this was one problem I hope to move on from.

I have thus far thought up three options of how to go about this:

• Merging objects and adding geometry to curve the edges of the room. However, this in theory would then create a perfect curve. Also unrealistic.

• So, the next option would be to use Mudbox to create rubble, bumps and imperfections. This would either entail sculpting and exporting a displacement map (and dealing with that in Maya), or re-importing the object with high subdivision levels - brave on the processing front.

• The third, and most convincing, option is to use photographs from the environment to inform the camera angles and composite the edges of the room into the render. This of course is up for testing to see if it can be roto'd and sync'd into the environment. But, I believe at this stage it will consist of making sure the textures and the compositing edge elements are similar and therefore easier to composite together. This of course fits much better into my Programme of Study which is based around high-end compositing.

3D Tracking Test

Up until now I have tracked shots with 2D (x and y) movements. Today, I have conducted a 3D tracking test (x, y and z) on the main handheld dolly shot.

This is because I see 3D tracking as a step up from 2D tracking and so I wanted to test my compositing workflow tracked to the Tracker node before the CameraTracker node.

In 2D tracking, you can manually pick one point that you think will track well. Then you copy the translate values to a Transform node on the trackable elements. In 3D tracking, the system picks the points. For this track I chose initially 300 tracking points. This is because the camera moves a real lot. After this, basically, you can export the camera information for rendering 3D elements to composite in 2D pipeline; composite 3D elements within Nuke; or composite 2D elements onto 'Cards' which are connected to the 'camera' in your 3D 'scene'... This give you an idea why I see 3D tracking as a progression up from 2D.

I have placed this 3D sphere into the plate:

(Bottom right)

The majority of work on this shot will consist of 2D roto cleanup and possible additions of subtle atmosphere and fog which are of 2D stock footage. So, I will see what this entails in regard to things looking correct or flat and odd.

Here is the track so far:

I am very pleased with the solve. I have achieved a low Solve Error and it looks good to the (my) eye. It follows the many movements of the camera convincingly. I am quite impressed by the Nuke workflow and output here.

More on this later.

Sh040C Update

Here is an update on Sh040C.

I have cleaned it yesterday (by removing the rope and preparing the plate) but today I begun to add in smoke. I used my slower edited footage of atmosphere.

This is how it looks once the atmosphere is positioned, roto'd, layered and tracked:

However, this is not the look I am after because it looks very flat, slapped on top of the image, right in front of us - in our faces. POW.

So, I want to create the effect in which the atmosphere is wrapped around and intertwined with the ladder on the bottom third of the frame.

Therefore, I have to roto the ladder and composite it back on top:

I found the most effect way, instead of tracking the roto, was to simply animate the roto seen as it moves along a 2D track anyway (it is a very flat movement.) This meant that no unnecessary areas of the character were appearing on top of the ladder or other areas - creepy!

Even though this seems like a cool idea, it doesn't quite yet create the look as it doesn't make the smoke surround the ladder, it puts it behind the ladder. Therefore we need to see some smoke on top of the ladder. The best thing to do here is to take the merge of the smoke which is flat on top of the ladder and roto whispy shapes and areas of the smoke and then composite this again on top of the ladder roto.

This is the effect you are left with:

And this is the shot so far with some more colour correction testing:

What I wanted to create here was to darken it and bring out that contrast, but also not to lose the detail of the face within the darkness. In order to do this I lowered the gain of the whole piece but upped the gain of the shadows. Another notable tweak was that I also upped the gain of the highlights for more contrast and gave it a white cold and dreamy softness with a glow node. This adds to the feeling of the scene. And, this is just still in test mode and not final. I have not yet had a look at it next to the rest of the shots to see if there are any consistency issues, for example...

Initial Scene Preparations

First thing's first. Well, actually, it is calculating lens distortion, removing grain and motion tracking and such.

Here are a few examples of my initial preparations for a shot and how this applies to the final touches of Sh040C:

When using lens distortion, there are a few techniques. I prefer using either of two options: Nuke can automatically calculate - in a similar fashion to tracking - the lens distortion. The other is drawing points along straight elements in the scene and that will calculate if these elements bend due to lens distortion. I first of all use the former and use the latter to test if the lines are straight. If this does not work, I opt for the latter and draw the points manually along the lines.

This is an example for the lens distortion process:

This is the footage with lens distortion from the camera.

I have now applied the 'image analysis' and drawn a line with 'line analysis' in order to check it is has worked and the straight lines are indeed straight. They are.

Looking at this line on top of the original lens distorted footage, you can see that there is now a gradual distance from the line in the corners of the frame. This clarifies that the lens distortion has worked.

It is also important to clean out any noise from your footage if you are going to composite anything into it which may have no or different types and levels of grain.

Here is an example of removing grain in this shot:

Looking again at the original footage there is obvious noise/grain.

For comparison, here is the image once the Denoise node has been applied to the footage.

After the compositing of undistorted and de-grained footage/elements, it is important to re-apply the lens distortion:

Before. Notice the distance between the image and the end of the frame at either side.

After. Notice that gap filled and the rest of distortion now visible.

After. Now all of the smoke (subtle) have the same lens distortion to the originally shot footage.

The lens distortion makes everything like it should again and fills the frame correctly, as well as draw attention to any elements in the framing which the director designed or the DoP chose to bring out in the shoot, along with the elements that have been composited (I believe that it is good to place things in the composite considering the later lens distortion to make sure it looks right.)

And, re-graining:

By sourcing a flat area of the original plate (such as the concrete), using the F_Regrain node, it gives me an idea of grain. But, it always seems to require changing the grain size and amount by looking at the original footage again to use the eye to match it as best as possible.

After.

This is my process so far, regarding the first few cases of using the very effective method.

Sh041 Update

Regarding the previous post on this shot, said issues have been altered.

Some other particular details have been created, for example, the shadow under the additional 'conjoined' extractor fan, the repainting of the wall artefact behind the chosen re-location and a shadow on the fan as this particular fan would indeed be falling gradually out of the highly lit area.

These effects are achieved by rotoscoping, either by drawing bezier shapes, colour correcting this and compositing it on top; or cloning and painting artefacts out. Luckily this shot is very, very still and so requires no tracking whatsoever.

This is the original shot plate for comparison:

Cleanup in progress

Using Nuke's rotoPaint node, one can remove elements such as unwanted rope from their shots by using fairly simple digital painting techniques. For example, the clone tool, smear, blur and painting colour.

It is what is called 'invisible work' as it is not meant to be known to ever exist in the first place. After all, filmmaking is trickery and often staged and faked. Such techniques are to make things either more appealing or more realistic. Another good example is the light I removed in Sh041 (visible in this post.)

Here is an example of a 'Conjure' shot in progress:

I have also toyed with some colour correction, which needs some work, but for me it looks more like what I want from the shot rather than the warm image above it.

Of course it isn't as simple as painting one frame if the camera and other elements are moving. One has to track the painting into the footage, and any other painting done for areas which are out of frame in the beginning, as well as roto anything which is in front of the painted-out element. The character in this case moves on top of the rope. So he needs to be rotoscoped:

If I had left this shot without roto'ing the character and compositing this back on top of the repainting:

The area of painted wall would be on top of the arm as if a part of his arm was invisible or missing:

This is then the rotoscoped arm over the repainting:

So, this is what I have by this stage:

Hopefully this stands to demonstrate some of my compositing and cleanup processes I am using shot to shot.