


We need a FrameHold node to prevent camera moving automatically for perspective projection adjustments.
Overscan in ScanlineRender node is helpful for rendering extra space around the format range, avoid any artefacts.


We need a FrameHold node to prevent camera moving automatically for perspective projection adjustments.
Overscan in ScanlineRender node is helpful for rendering extra space around the format range, avoid any artefacts.
Smart vector is a powerful and expensive node in nuke for difficult tracking of face.
Sometimes we may use a combined method to do the tracking with the help of swith node.
Mocha is also a good plugin for tracking.
Smart vectors are usually used in beauty work:
We do all the beauty work in a reference frame.
The first step is always denosing the plate. In this case we use denoise node as well as median node, they are both heavy-working nodes and median node is a kind of filter node.
We clean up the plate afterwards. With rotopaint node, we clone for large flaws, remember to set the output in RGBA-alpha channel.
Also we have a excellent gizmo node called VanityFix, it works perfectly with tiny and high-intesity flaws.
We union those alpha outputs and premult them.
SmartVector node may take some time to run, depending on machines. The best extension for vectors file is .exr.
Then we will use VectorDistort node to track the face directly, and regrain to match the original plate.
The final step will be merge the beauty work with the original plate, we can merge it twice, the last one is used to recover some imperfection to make the final plate more believable.
When tracking a plane in a perspective movement, I tried to make a loop on the monitor.
It is noticeable that the output from cornerpin2D node contains an alpha channel. If we want to add on any elements, we need to premult it before sending it to the next cornerpin2D, otherwise alpha channel will double in the matching region.
This time, I tried to clean-up the plate firstly, and place a transparent logo image on the headphone. It is quite tricky to remove the original logo, because the existing reflection light is changing.
I tried to re-roto my first rotoscoping, and use the grade node to match the colour with the background plate.
I tried to place a logo on the headphone this time:
Lift performs in an opposite way to blackpoint, wheras gain performs in an opposite way to whitepoint.
Multiply performs in the same way as gain.
Source:
Firstly, I set a physical sky in maya, and the scene seemed too dark, so I just enhanced the light intensity later.
I also tried different backgrounds downloaded from the website. HDR maps need to be connected to Ai SkyDomeLight. For the following rendering, I increased Samples to be 4, it would take longer but could get a better quality result.
When I felt satisfied with the rendering for the still image, I created a camera and its aim in order to render a sequence of frames.
Here are render settings that I used:
When finishing sculpting and painting in Mudbox, I extracted texture maps under UVs & Maps. There are multiple types of maps to choose, I chose vector displacement map here because it contains more information than displacement map, so it might perform better when rendering with Arnold in Maya. Normal maps are commonly used in games.
We have some options for vector space. Because I just need to render it using Arnold, I just chose Absolute Tangent, it is a choice with the most flexible effect.
If we want to use the extracted maps as stamps or stencils used in Mudbox, Relative Tangent will be necessary.
World vector space is used for the undeformed and still objects such as environment. Object vector space is used for the undeformed objects usually.
Whatever the vector space is, we need to select the same one in Arnold material settings.
It is important to use 32-Bit for displacement maps.
Remember to tick on smooth target UVs, otherwise the rendering result will show errors.
The file which was sent from Mudbox was set in phong material as default. Therefore, we need to assign an Arnold material to the material, such as aiStandardSurface. We can edited the attributes and insert images in Hypershade window by using node graphs. For displacement maps, it is important to use raw colorspace without filters, as this type of images need to transfer accurate information.
Set the subdivision type as catclark with suitable iterations will improve the rendering result. A high number of iterations will result in longer export/render times.
I also tried to render it with a displacement map, here is the result:
For the background board, I downloaded the texture for website and used the normal map instead of displacement map or bump map.