Image: Compositing & Masks#

Blending images and building masks.

10 nodes.

Back to the reference index

Contents#

• COP Combine Normals (PCG) — Combines a base normal map and a detail normal map into one normal map.
• COP Compare Mask (PCG) — Compares two layers (or a layer and a constant) per pixel and outputs a 0..1 mask.
• COP Composite (PCG) — Composites the foreground input over the background input with a choice of blend modes.
• COP Drop Shadow (PCG) — Builds a drop shadow from an alpha mask by offsetting, blurring and tinting it behind the original image.
• COP Layer Combine (PCG) — Combines the same layer from up to six inputs into one image, pixel by pixel.
• COP Logic (PCG) — Compares layers or combines them with boolean operations, producing a 0/1 mask.
• COP Mask Combine (PCG) — Combines several masks into one, using a simple operation or a Houdini-style stack of blend modes.
• COP Premultiply (PCG) — Multiplies RGB by alpha, or divides it back out.
• COP Select (PCG) — Compares two layers per pixel and picks one, or blends them smoothly by the difference.
• COP Wipe (PCG) — Transitions from image A to image B with a linear, radial, dissolve or crossfade wipe.

Nodes#

COP Combine Normals (PCG)#

PCG_COPCombineNormals · category PCG/cop · Houdini: Combine Normals (COP)

Combines a base normal map and a detail normal map into one normal map.

Both inputs must be signed normals (-1..1). The base layer is read from cop; the detail layer from the detail input, or from the same image if detail is not connected. Methods:

• rnm (default): Reoriented Normal Mapping. Rotates the detail into the base's frame, so detail on steep slopes follows the slope instead of being flattened.
• whiteout: adds xy, multiplies z. Cheap, but detail is weaker on steep slopes.
• udn: adds xy, takes z from the base. Even cheaper.
• linear: adds and renormalizes. Kept for comparison only: even flat detail tilts the base towards +z and weakens slopes.

With flat detail, rnm, whiteout and udn return the base unchanged (linear does not); with a flat base, rnm returns the detail unchanged. The result is always unit length. Amount scales the detail (0 means no detail).

Inputs

Outputs

Parameters

COP Compare Mask (PCG)#

PCG_COPCompareMask · category PCG/cop · Houdini: Compare (COP)

Compares two layers (or a layer and a constant) per pixel and outputs a 0..1 mask.

Compares A with B per pixel using the True If condition and writes a single-channel 0..1 mask. B is Layer B, or the Comparison Value when Layer B is empty. For multi-channel layers, Components decides whether all channels (minimum) or any channel (maximum) must pass.

Tolerance is two-sided for the equality tests (equal, not_equal) and one-sided for the strict and non-strict inequalities. Pixels that pass the comparison are always 1; Falloff shapes the pixels that fail it but lie within the tolerance: none cuts hard at the tolerance edge, linear ramps from 1 down to 0, elendt uses a smooth S-shaped curve (implemented here as smoothstep, so intermediate values differ slightly from Houdini). not_equal is the complement of equal.

To measure how different two images are (SSIM, PSNR) use COP Compare.

Inputs

Outputs

Parameters

Compared with Houdini

Corresponds to Houdini's Compare node.

COP Composite (PCG)#

PCG_COPComposite · category PCG/cop · Houdini: Blend (COP)

Composites the foreground input over the background input with a choice of blend modes.

over, atop, xor and under are Porter-Duff operators and compute a new alpha: C = (Cf*af + Cb*ab*(1-af)) / a, a = af + ab*(1-af) for over. They work on non-premultiplied alpha, like ComfyUI images. All other modes (add, multiply, screen and so on) are component-wise functions of foreground and background; opacity and mask then interpolate between the background and the blended result. Treating over as an alpha interpolation would give the wrong alpha when both layers are semi-transparent.

If resolutions differ, the foreground is resampled to the background. Only single-channel layers are broadcast automatically (for example a mask onto RGB); a 3- versus 4-channel mismatch is not guessed. Values are never clamped, so add can give 1.7 and HDR, SDF and normal data survive; clamping only happens at COP Export.

The mask input is multiplied with Opacity when connected.

Inputs

Outputs

Parameters

Compared with Houdini

Compared with Houdini's Blend, Under, Hypot, Exclusion, HardLight, Dodge and Burn were added; Houdini's Blend is replace here, and Maximum / Minimum are lighten / darken. The Sharpen mode and the Hue, Saturation, Luminosity and Color modes are not implemented. Space, Border, Filter, Clip by Data Window and Crop to Content are not offered because images have no data window.

Tips

• Keep HDR data in the chain and clamp only at the end (COP Export or COP Clamp).

COP Drop Shadow (PCG)#

PCG_COPDropShadow · category PCG/cop · Houdini: Drop Shadow (COP)

Builds a drop shadow from an alpha mask by offsetting, blurring and tinting it behind the original image.

Takes the alpha (the 4th channel of layer, or a separate alpha layer), shifts it by an offset, blurs it, colors it and places it underneath the original image. The result is RGBA.

The node errors out if there is no alpha at all; it does not guess one.

Inputs

Outputs

Parameters

COP Layer Combine (PCG)#

PCG_COPLayerCombine · category PCG/cop · Houdini: Layer Merge (COP)

Combines the same layer from up to six inputs into one image, pixel by pixel.

Reads the chosen layer from every connected input and combines them per pixel with Add, Average, Multiply, Min, Max or Over. Resolution follows the first connected input and the others are resampled to it; single-channel layers are broadcast to the channel count of the others.

Over places earlier inputs on top of later ones and only makes sense for RGBA; for non-RGBA layers the last input is copied. Houdini accepts any number of inputs; here there are six sockets, so chain two nodes if you need more.

This is not COP Layer Merge (PCG), which moves layers from one image into another without doing any pixel math.

Inputs

Outputs

Parameters

Compared with Houdini

Corresponds to Houdini's Layer Merge COP (the one documented as cop/average). Space, Border, Filter, Clip by Data Window and Crop to Content are not implemented because they depend on data windows.

COP Logic (PCG)#

PCG_COPLogic · category PCG/cop · Houdini: Logic (COP)

Compares layers or combines them with boolean operations, producing a 0/1 mask.

Input A is a layer name; Input B is either a layer name or a constant number. Comparison operations (A > B, A >= B, A < B, A <= B, A = B, A != B) give 1 where true. Boolean operations (AND, OR, XOR, NOT A) treat any value above 0.5 as true. Equality uses Tolerance, since exact float equality almost never holds. For multichannel layers the result is 1 only where all channels satisfy the test. The output is a single channel.

Inputs

Outputs

Parameters

COP Mask Combine (PCG)#

PCG_COPMaskCombine · category PCG/cop · Houdini: Mask Combine (COP)

Combines several masks into one, using a simple operation or a Houdini-style stack of blend modes.

Only the first channel of each layer is used. Simple operations on the listed Mask Layers: Union is the maximum, Intersect the minimum, Add sums, Multiply multiplies, Subtract takes the first minus all others, Difference accumulates |a - b|, Screen is the soft union 1 - prod(1 - m), and Average takes the mean.

Per-Layer Blend Modes (Houdini) works like Houdini's node: the result starts at Starting Value and the rows of the Mask Layers stack are applied from top to bottom, each as result = lerp(result, blend(result, layer), amount). Each row has the form layer[.channel] mode amount; the channel is r g b a or x y z w (default first channel), mode defaults to blend and amount to 1, and a row starting with # is disabled. Rows may also be separated with ;. This mode ignores the simple Mask Layers list.

Post-processing is applied in this order: Fold (absolute value), Complement (1 - x), Clamp Minimum, Clamp Maximum, then Clamp 0-1.

Inputs

Outputs

Parameters

Post Processing

Compared with Houdini

The Per-Layer Blend Modes option follows Houdini's multiparm; here it is written as multi-line text. Per-layer Channel Mix, Light Direction, Blur and Layer Remap, and the post-process Remap, Amplitude, Center, Contrast, Bias, Gain and Gamma are not implemented; add separate COP nodes before or after if needed.

COP Premultiply (PCG)#

PCG_COPPremultiply · category PCG/cop · Houdini: Premultiply (COP)

Multiplies RGB by alpha, or divides it back out.

Premultiply scales the colour channels by alpha. Unpremultiply divides them back; where alpha is 0 the result is 0 instead of NaN.

Inputs

Outputs

Parameters

COP Select (PCG)#

PCG_COPSelect · category PCG/cop · Houdini: Compare and Blend (COP)

Compares two layers per pixel and picks one, or blends them smoothly by the difference.

For each pixel, chooses A or B based on the comparison: A greater, A less, brighter or darker by luminance, closer to a Value, or a key layer exceeding a Value. With Softness above 0 the choice becomes a smooth blend based on the difference. Optionally the selection mask is also written.

Inputs

Outputs

Parameters

COP Wipe (PCG)#

PCG_COPWipe · category PCG/cop · Houdini: Wipe (COP)

Transitions from image A to image B with a linear, radial, dissolve or crossfade wipe.

The transition amount T runs from 0 (only A) to 1 (only B). At exactly 0 and 1 the result is bit-identical to A and B. The modes:

• Linear: a soft edge sweeps across the image in the direction of Angle.
• Radial: a circle grows from Center.
• Dissolve: each pixel switches at its own random threshold (seeded); larger softness blurs the threshold.
• Crossfade: a plain blend.

If B has a different resolution than A it is resampled to A first; only the first channels shared by both layers are used. The result takes the resolution and other layers of A.

Inputs

Outputs

Parameters