| On this page |
Old network
As of Houdini 20.5, use Copernicus nodes instead of Compositing nodes. Though both networks still exist, the Compositing network is now designated as COP Network - Old. The Compositing network and its nodes will be deprecated and then removed in a future Houdini release.
Houdini includes a network based, procedural compositor. As in other areas of Houdini, to do compositing in Houdini you load or create pixel data with “generator” nodes, and then pipe the data through a series of “filter” nodes to modify the data.
A compositing node (COP) manipulates image sequences (internally, however, it generally processes one plane at a time) and outputs an image sequence.
Organization and data types ¶
The Houdini compositor works with sequences of images. An image is made up of planes. Planes are two dimensional maps of pixels, which can store 1-4 channels or arrays of data.
Sequence
A sequence is a series of images, with a start frame and a length.
All images in a sequence have the same dimensions (that is, you cannot have one frame at 800×600 pixels and another frame at 1024×768 pixels in the same sequence).
A sequence can specify how the data behaves outside the sequence’s frame range using pre- and post extend conditions. For example, you can hold the first/last frame, just show a black frame, cycle the sequence, or mirror (A.K.A. ping-pong) the sequence back and forth.
The properties of a sequence are constant over time; you cannot animate them.
Image
An image is a single frame within a sequence. The image can be made up of many planes, each storing different types of information, such as color.
Plane
The Houdini compositor supports deep rasters. This means an image can hold many layers of information for each pixel (up to 1000 in the Houdini compositor). Each layer of information is called a plane. In addition to the traditional RGBA color information plane, you can store information such as depth, normals, or mattes. See “standard planes” below.
Pixel
A plane is a two-dimensional map of pixels. Each pixel can store 1-4 channels (see below), or an array of 1-4096 elements.
When you store arrays, you can refer to the elements of the
array using ‹plane›[‹index›]. Each element can contain
different values, but all elements of the array have the same
data type and component structure (see components below). This
is commonly used to represent cubic environment maps: store
6-element arrays in the color (C) plane to represent the 6
faces of the cube.
Because pixels can store arrays of data, you could represent a plane of 4-dimensional matrices (Matrix4) as plane of 4-element arrays, where the arrays contain elements with 4 32-bit floating point channels (Vector4).
Channels
A channel is like sub-data of a pixel. For example, on the color
plane (C, see “standard planes” below), each pixel has red
(r), green (g), and blue (b) channels.
An element with 1 channel is called scalar. An element with 2 channels is called 2-channel. An element with 3 channels is called Vector3. An element with 4 channels is called Vector4.
You can refer to channels of a plane using
‹plane›.‹channel› . For example, C.r means the red (r)
channel of the color (C) plane.
Channels can hold one of five different types of data:
8 bit integer
Can store integer values 0-255. As smallest integer format, uses less memory, faster to process. High level of quantization, OK for video, but very lossy.
16 bit integer
Can store integer values 0-65535. Traditional representation of color for film. Some optimizations over other formats, less quantization. Uses twice as much space as 8-bit integer. Still not a good HDR format (even with black/white points) because it is linear.
32 bit integer
Can store integer values 0-232. Not intended to store color information: good for storing object or point IDs, particle lifetimes, etc. Can be used to uniquely identify millions of values, but high memory usage and slow to process.
16 bit float
Can store fractional values from -65535.0 to 65535.0. Gives good resolution between 0 (black) and 1 (white). Quantization increases with magnitude, which is fine for HDR images. Can represent a very high HDR contrast ratio. Smaller memory usage than 32-bit floating point. Good at representing bump maps and normal maps with good precision.
32 bit float
Can store fractional values from -1e32 to 1e32. Excellent resolution for most compositing applications. Uses a lot of memory space and bandwidth.
The integer formats can optionally have white and black points set (see black point and white point ). By default, the white and black points are the 0 and the maximum value of the format (8 bit - 0, 255; 16 bit - 0, 65535; 32 bit - 0, 2 billion).
Standard planes ¶
The following planes are recognized by the Houdini compositor.
Color (C)
Image color, represented by three channels, r, g, and b.
Alpha (A)
Image alpha, 1 channel.
Mask (M)
Operation mask, which acts as a stencil for a filter operation. 1 or 3 channels.
Depth (Z)
Z depth, 1 32 bit floating point channel.
Point (P)
3D camera space point position, 3 32 bit floating point channels
(x,y,z)
Normal (N)
Object point normals, 3 32 bit floating point channels (x,y,z)
Bump (B)
Bump map, 2 32 bit floating point channels (u,v)
Velocity (V)
Point Velocity map, 3 32 bit floating point channels (x,y,z)
Luminance (L)
1 single channel.
Other planes may be generated by mantra, such as any exported shader variables. These may be used by the compositor, but it does not recognize them automatically (similarly to spare channels).
The Houdini compositor is multi-threaded (supporting up to 8 threads) for performance, and takes advantage of SSE and MMX instructions on processors that support them.