Production Line & Breakdown from Liam Major on Vimeo
The latest humanoid machine prototype is being tested on the production line for the first time. Basic response tests, such as open mouth and rotate head, are being performed. Of course some bugs still need to be fixed.
Figure I & II: Raw plate & final composite
The brief was to demonstrate nodal compositing ability. This gave me full creative control but I planned the project to involve many advance compositing techniques for maximum marks. My technical concept was to replace parts on a human body with 3D parts, tracked into place. I felt a little weak when it came to keying, so decided to have it all filmed on green screen. Take the bull by the horns and all that. I also thought it’d be an interesting shot to have the camera static and locked onto the machine, which moved down a production line so the environment moved around the subject.
Process
Previsuals
To get an idea of timing, composition and what needed to be done I created a previs animation in After Effects using stock images. It gave me some direction and visual structure.
Figure III & IV: Previsual animation created within After Effects
Filming
I filmed on an green infinity curve, created by a group of final year BDes students at Bolton University SFX. My actor, Dan Twose, performed different movements at different speeds to give me extra control. I also shot without tracking markers and basic head movement as an alternative in case the 3D track failed.
I placed some markers on the actors head. Masking tape with pencil crosses (very technical stuff). Of course I could have just drawn on his head, but he may have not been so pleased. I deliberately chose an actor with no hair to avoid complications, what with such a short time. The marks were placed exactly where I wanted the CG to be. This removed the need of painting them out.
Match Move
With the markers I tracked the object movement (not the camera, which was locked) with PFTrack. The track wasn’t perfect which I put down to the movement being mostly rotation. There is very little sense of depth (without parallax), which made it difficult. Some frames turned out to cause twitches. To solve this, I delete these keyframes and allowed Maya to tween the accurate frames.
Figure V: Object match moved with PFTrack
I exported the track data to Maya. Rather than applying this data onto the 3D models themselves, I applied them to a group. All the models were then kept within this group. This gave them the tracked movement but still preserved independent control over their rotation, translation and scale.
3D
The modelling consisted of basic machinary/type writer styled objects. They were shaded (Mia_X_Passes) to appear like new, clean chrome straight out of the factory. I modelled a backplate for the machinery parts that roughly followed the shape of the inner head, shown below.
Figure VI: Final model with Maya & Mental Ray
I placed a proxy head within Maya to match the position of the actors head. This allowed me have reference to where the CG parts where placed. It also enabled me to place machinery outside the actors head too, which I did. Further more it allowed me to render shadows of the 3D on to the real head.
I used the head for another purpose too; generating a matte for the hole in the head. I used a surface shader, that don’t react with light, and filled it black with a white section for the hole (done with the ‘paint’ tool in Maya).
Figure VII: Proxy head screen shot from Maya
All the animations were created using script to save time. I created lots of movement to keep the machinery interesting. ‘rotation.X=frame’, for example, caused a rotation on the X axis based on the frame number. Some of the components were animated by hand to match with the actors movements.
Because the CG machinery was highly reflective a reflection rig was needed to create realistic reflections. I created a large plane and textured it with the same environment created within Fusion (which had been rendered without the actor and before colour corrections and blur). This caused the reflections to move with correctly.
It was rendered using Mental Ray with linear work flow in passes. I rendered it at 1065 x 600 pixels, even though the footage was 1080p. Because film isn’t perfectly sharp and clean like CG renders, it allowed me to render at a lower resolution and scale it up without noticeable loss of quality. There was no need to blur the render in post either.
Figure VIII: Render passes
CG Environment
I created the environment within Fusion. Most (or all I’d presume) compositing packages include some sort of 3D (or 2.5D, depending how you look at it) engine and when used correctly they can be very powerful. Of course, it’s not designed to replace a 3D package. I used it in this project to create parallax with basic camera movement. In this example, I managed to turn 10 images into a 3D environment.
Figure IX: Environment within Fusion
I used free textures from CG Textures. Most were applied to planes. I applied textures to cubes for the close up pillars. This created realistic perspective as they moved from right to left. All the objects were static and the camera moved through the scene. A slight wobble was applied to the Y axis position of the camera to make it feel less static and not completely locked onto the actor. This made the movement feel much more mechanical.
I put all the environments with similar z-depths through the same render nodes. The render node essentially flattens the 3D so it can be merged with 2D nodes, such as imported footage. This split each layer of 3D into its own flat sequence, which allowed me to place the footage below the CG foreground. The problem with this is I had four 3D environment layers that all needed the same camera. Although separate, the same 3D camera node was input into each layers scene. The beauty of nodes!
Figure X & XI: Before & after depth blur
Because each layer of the 3D scene was rendered separately, I had control over editing them separately. I applied blur to each layer, the blur amount depending on the distance from the camera. This transformed the scene in a positive way, demonstrated above.
Compositing
As mentioned, keying green screens was something I usually didn’t enjoy. Not until this project. The infinity curve gave me an instant great screen though. No creases, stitches, stains or various shades of green. For this reason, I was able to get an excellent matte using only 1 key. I checked for holes and crunching, but it was fine. I find it good practise, for many reasons, to split the key into its own flow and later in the flow multiply it to the original footage. Despill reduced some of the green channel caused by indirect illumination from the screen. The despill was subtle to avoid turning the footage magenta (compared against the raw plate to check).
For this project I used ’Ultra Keyer’. This key operates by calculating the different (subtracting) between the red, green and blue channels. As a consequence, the key can only key a primary colour and as red isn’t used (due to skin pigment) so the user is limited to green and blue. To test the key its best to analysis the RGB levels of the image. Ideally, there will be high contrast between the key colour (in this example green) against the other two colour channels (red and blue). If the screen turns out a different colour, say cyan, it maybe best to use a chroma key.
I applied a light wrap and edge blur to the keyed actor and CG. I won’t go into much detail because I’ve mention the process in detail in my other projects. Essentially it helped blend the pixel values around the edges of the objects. Film grain was applied after using the same mattes.
Because I rendered the 3D as passes, I recombined them by adding them (light is additive). They were multiplied (using a boolean) by the rendered head hole matte. Now, I also rendered off a multimatte purely for colour correcting parts of the machine depending on the depth within the head. For example, close machinery was red and machinery at the back of the head was green. I used these mattes to colour correct the insides without having to render the machinery as 3 image sequences.
Figure XII: Head mattes after various mathematical processes
At this stage is was apparent there needed to be sort of edge with lighting on the skin around the hole in the head. I blurred the head matte and then subtracted it by itself (before the blur). This gave me a feathered matte for the outside of the hole (amazing what you can achieve with basic mathematics). I made multiple versions, by changing the blur amount, and input them as effect masks into a colour corrector node. I changed the colour to make the flesh look burned and bloody.
One of these mattes was multiplied by a linear black and white gradient (making it fall off) following the direction of the light sourse. I used this to increase the highlights. The same gradient was inverted (which gave an opposite matte) and input it into another colour corrector to create shadows. The pair created a fake lighting system. The beauty of this is both sides are effected by the same gradient, meaning I could manually change both highlight and shadow together. In essence, they were always opposite.
I used colour grading for three purposes; legality, composition & mood.
By legality, I mean making sure the colour values are safe to broadcast. Vectorscopes analyse the saturation of the sequence and waveforms the brightness. I like to keep the colours safe by keeping the brightness between 0.2 – 0.8 (rather than 0 – 1). I also made sure the darkest points on my actor were similar to the darkest points in the environment.
Figure XIII & XIV: Vectorscope and Waveform of the final scene
I used colour to effect the composition of the shot. The eye is naturally drawn to areas of high saturation, contrast or brightness. A good way to see where the audience will be attracted to is to blur the image, say around 20 pixels, and see what spots stand out (or if you’re anything like me, just take your glasses off). Back to this project, I darkened and slightly desaturated the outer edges of the shot. The windows caused very bright spots so I lowered the gain. I also, slightly, increased the gamma in the center (using an iverted matte) just to lift the saturation & brightness in the midtones.
Figure XV: Before colour correction
Figure XVI: After colour correction
This project only had 1 shot, so creating a unified scene over many shots with colour correction wasn’t an issue. However, I still used it to create a mood. I felt this scene needed to be dirty, unhealthy and slightly threatening. I reduced the blue and red channel to give it a slightly yellow tint. I also to increased the green, which over all gave an unhealthy vibe. There were many tweaks, but that’s the general principle behind it. Gradients are great for colour corrections and can make the colours a lot more interesting. You may notice, in the image above, the green/yellow haze that falls off out of the bottom left corner.
Figure XVII: Final Fusion flow
Probably a slightly more complicated flow, mainly due to the environment. Hopefully this breakdown will have made it simple to understand. Comment below any questions. Cheers.
- Credit
- Liam Major Filming, Modelling, Texturing, Dynamics, Tracking & Compositing
- Daniel Twose Actor
- Software
- Autodesk Maya 3D modelling, texturing & animation
- Eyeon Fusion Compositing & CG environment
- Pixel Farm PFTrack Object match move
- Client
- Bolton University Advance Visualisation assignment
- Duration
- 4 weeks (during 2 other projects)
Driebrapypope
24 Jul, 12 o′clock
why not…
Liam Major
24 Jul, 1 o′clock
.. exactly