Convert#

Conversions between primitive types, volumes and other representations.

7 nodes.

Back to the reference index

Contents#

• Convert Line (PCG) — Converts polygon edges into line primitives.
• File Merge (PCG) — Reads a numbered series of mesh files and merges them into one geometry.
• From JSON (PCG) — Builds points, boxes or curves, with attributes and groups, from a JSON text.
• Geo to JSON (PCG) — Exports attributes of points, primitives or the detail as a JSON text.
• Import Mesh (PCG) — Reads a mesh or point cloud file (OBJ, PLY, STL, GLB / glTF, LAS / LAZ) in pure Python.
• Trace (PCG) — Traces the iso-contours of an image into polyline curves.
• Volume Rasterize Attributes (PCG) — Splats points into volumes: a density fog plus a kernel-weighted average of chosen point attributes.

Nodes#

Convert Line (PCG)#

PCG_ConvertLine · category PCG/convert · Houdini: Convert Line (SOP)

Converts polygon edges into line primitives.

Every edge of the selected primitives becomes a two-point polyline. Shared edges are emitted once, deduplicated by point pair; emitting per face corner would double every interior edge, which looks identical as a wireframe but gives a cloth or spring solver two overlapping springs with doubled stiffness. Primitive attributes are taken from the first primitive that uses the edge; vertex attributes come from the corresponding half-edge.

With Connect Path on, edges are chained into the longest polylines possible instead of two-point segments: a walk continues through points of degree 2 and stops at points of any other degree. This is what makes Resample, Carve and Sweep treat the result as curves. Chains start at endpoints first and closed loops are handled afterwards, so an open chain is never split in the middle. When Compute Length is on the edge length is written to a primitive attribute.

Inputs

Outputs

Parameters

File Merge (PCG)#

PCG_FileMerge · category PCG/convert · Houdini: File Merge (SOP)

Reads a numbered series of mesh files and merges them into one geometry.

The $SLICE token in the file name (or the variable named by Merge Variable, also accepted as ${SLICE}) is replaced in turn by each number of the Merge Range (start, end inclusive, increment; a negative increment counts down). The files are read in order and merged. .npz files are read as this plugin's geometry cache; other files go through the same format sniffing as Import Mesh with its default settings.

With Delete Points from Other Files, points whose slice point attribute differs from the current number are removed, which is useful for distributed simulations where each file also holds particles of other slices. Files without that attribute are left untouched. With Missing File set to No Geometry, absent files are skipped and listed in the report.

Outputs

Parameters

Compared with Houdini

Houdini's Load and Cache Length parameters are not present; ComfyUI caches by node inputs.

From JSON (PCG)#

PCG_FromJSON · category PCG/convert

Builds points, boxes or curves, with attributes and groups, from a JSON text.

The entry point for LLM / VLM driven layouts: the model emits structure (positions, bounding boxes, labels, categories) and this node turns it into geometry that downstream nodes (Copy to Points, PolyExtrude, Blast, Switch) can use.

Parsing is tolerant, because models rarely return clean JSON. Four stages are tried in order: parse directly, strip a markdown code fence, extract the first balanced [...] / {...} block, remove trailing commas. The stage that succeeded is reported in info, so a silent repair never goes unnoticed.

Modes: points makes one point per record (pos, P or position); boxes makes one rectangle (4 numbers) or box (6 numbers) per record, e.g. detection boxes or room layouts (box, then PolyExtrude gives walls); curves makes one polyline per record (points or path). Record keys are mapped to canonical names: pos / position / translate become P, color / Cd become Cd (hex strings such as "#ff8800" are understood when Hex Colors is on) and label / type / tag become name. Other keys are kept as they are: numbers become float attributes, fixed-length arrays become vectors, strings become string attributes. Attributes land on points in points mode and on primitives in the other modes.

Group By turns every distinct value of an attribute (usually name) into a group, so a classification from the model becomes usable groups.

Coordinates follow one rule for all modes. Two-component coordinates with Normalized on map 0..1 onto a Size sized area on the chosen plane (default xz ground plane); with Normalized off they are world units on that plane. Both are then offset by Center. Three-component coordinates are world coordinates plus Center. Flip V is for image space, where y points down.

Outputs

Parameters

Compared with Houdini

There is no single Houdini equivalent; the closest is building points and curves with Add / Curve from external data.

Geo to JSON (PCG)#

PCG_GeoToJSON · category PCG/convert

Exports attributes of points, primitives or the detail as a JSON text.

Meant to show a language model the current layout so that it can propose a matching next step. It uses the same key conventions as From JSON, so geometry to JSON, edit by the model, and back through From JSON is a closed loop.

Only attributes are exported, never topology: thousands of vertex indices would only use up the model's context. Use Export Mesh or File Cache for full geometry. Vector attributes become arrays, scalars become numbers and strings stay strings.

When Limit truncates the output, the result becomes an object with items, truncated: true and total, so the model knows it does not see everything.

Inputs

Outputs

Parameters

Import Mesh (PCG)#

PCG_ImportMesh · category PCG/convert · Houdini: File (SOP), Lidar Import (SOP)

Reads a mesh or point cloud file (OBJ, PLY, STL, GLB / glTF, LAS / LAZ) in pure Python.

Intended for the output of image-to-3D models, which usually have the right shape but poor topology: triangle soup, duplicate points, no UVs, uncontrolled face counts. A typical repair chain is Import Mesh, Fuse, Facet (recompute normals), PolyReduce, UV AutoSeams, UV Flatten, UV Layout, Export Mesh.

Nothing is triangulated on import: OBJ n-gons are preserved, which PolyReduce's quad preservation and Dissolve's coplanar merging need.

Several parameters fail silently when wrong. Weld: STL stores every triangle with its own three points, so without welding the point count triples and nothing is connected; Subdivide, PolyReduce, Connectivity and Normal cusp then quietly do nothing. auto welds STL only. Up Axis is never guessed: glTF is Y-up, OBJ / PLY / STL have no convention and Blender often exports Z-up, so you choose. Format auto looks at the file's magic bytes, not its extension; binary STL is recognised from its length (84 + 50 x triangle count) because its header often begins with "solid".

UVs are written to vertices, not points: OBJ vt indices are per corner (the same point can have different UVs on different faces, which is what a seam is), and glTF per-point UVs are expanded to vertices as well, matching the default class of all UV nodes in this plugin. A PLY without faces is read as a point cloud with every property kept as a point attribute, so 3D Gaussian Splatting files load directly (opacity, scale_*, rot_*, f_dc_*). With Splat Color on, the zero-order spherical harmonic coefficients become Cd using C = 0.282095 * f_dc + 0.5.

The mesh input accepts an in-memory mesh from another 3D plugin by duck typing: objects with .vertices / .faces, a dict with vertices and faces, or lists of those; torch tensors are converted. When it is connected, Filename is ignored, and Weld on welds while auto and off do not.

LAS / LAZ LiDAR point clouds: uncompressed .las (point formats 0-3 and 6-8) is read without any dependency; compressed .laz needs the optional laspy package (pip install "laspy[lazrs]"). Points get intensity, classification, return_number, number_of_returns, scan_angle, user_data, point_source_id, gps_time where present (Keep Extra Attributes), and Cd from RGB. Survey coordinates (hundreds of kilometres) do not fit float32 positions, so a cloud with any coordinate above 10 000 is moved to the rounded centre of its bounds and the offset is kept in the detail attribute las_origin (world = P + las_origin). LAS is Z-up: set Up Axis to Z-Up to Y-Up.

Inputs

Outputs

Parameters

Compared with Houdini

Combines the roles of Houdini's File SOP (formats, up axis, scale) and parts of Lidar Import (point cloud PLY); only OBJ, PLY, STL and GLB are supported.

Trace (PCG)#

PCG_Trace · category PCG/convert · Houdini: Trace (SOP)

Traces the iso-contours of an image into polyline curves.

The one step on the image to geometry path that brings back topology: Attribute from Map only gives values, while Trace produces curves that can be extruded, swept or used as cutters. Line art, silhouettes and masks from image models become editable geometry.

The algorithm is Marching Squares with two points handled carefully. Saddle cells are ambiguous (the two segments could be connected or separate), and a wrong choice changes the topology, so the average of the four corners decides, which is exact for a bilinear field. Orientation is consistent: the side above the threshold is on the left, so outer contours are counter-clockwise (positive signed area) and holes are clockwise (negative), which PolyFill and PolyExtrude rely on.

Simplify is measured in image pixels, not world units, because it means how far the curve may deviate from the image and should not change with output size. Closed loops start the simplification from the farthest pair of points so a circle does not keep a spike at the start.

Output primitive attributes are perimeter and area (signed, in world units; open curves get 0 area) and the point attribute uv, which records where on the image each point came from and can be used with Attribute from Map to fetch colors.

Inputs

Outputs

Parameters

Placement

Compared with Houdini

Resample Shapes, Boundary Color and the Transform page follow Houdini's Trace parameters. Houdini's Trace produces faces; here the output is closed or open polylines, and use PolyFill to get faces.

Volume Rasterize Attributes (PCG)#

PCG_VolumeRasterizeAttribs · category PCG/convert · Houdini: Volume Rasterize Attributes (SOP)

Splats points into volumes: a density fog plus a kernel-weighted average of chosen point attributes.

Each point is a sphere of radius pscale x Radius Scale (or Radius when pscale is off or missing) with kernel weight (1 - d/r)^2. The output holds a fog volume named density (the sum of kernel weights times the point's density attribute, or 1 if the point has none) and one volume per requested attribute holding the kernel-weighted average, not the sum: two overlapping red points stay red. Texels with zero total weight are 0. Vector attributes are split into scalar volumes named name.x, .y, .z (.w).

Voxels are cubes aligned to world-space multiples of Voxel Size; the extent is the point bounding box grown by the largest radius. An error is raised if the grid would exceed 300^3 voxels, or if no point has a positive radius.

Inputs

Outputs

Parameters