pcg

PCG (Procedural Content Generation) in UE 5.7 is production-ready and fully scriptable via the PCGPythonInterop plugin. No VibeUE service wrapper is needed — the Python API is complete.

⚠️ Read this before writing any code

These are the mistakes that cause infinite retry loops. Check each one before executing:

Do not call discover_python_class or discover_python_function on any PCG class — this skill contains everything you need. Calling discover bloats context and causes content filter failures.

1. Volume Sampler input pin is "Volume", not "In" — add_edge(input_node, 'In', sampler_node, 'Volume'). Using 'In' silently no-ops and produces 0 points.
2. voxel_size is a Vector, not a float — set_editor_property('voxel_size', unreal.Vector(200, 200, 200)). Passing a float raises a TypeError.
3. graph property on PCGComponent is protected — use pcg_comp.set_graph(graph), never set_editor_property('graph', ...).
4. After modifying a graph, always: save → reload → reassign → generate — skipping reload means the component runs the old cached version.
5. Check for existing nodes before adding — always inspect [type(n.get_settings()).__name__ for n in graph.nodes] first. Adding a node type that already exists creates duplicates that break wiring.
6. Always save AND reload after every change — call save_asset then reload_packages every time. Without the reload the PCG graph editor window won't update and users will think nothing happened.
7. Use Volume Sampler for empty levels, not Surface Sampler — Surface Sampler requires geometry to sample from. An empty level has no geometry, so Surface Sampler always produces 0 points. Use PCGVolumeSamplerSettings with unbounded=True for empty or sparse levels.
8. Never spawn test actors to verify meshes or materials — do not use spawn_actor_from_class to place test cubes. Verify by checking ISM instance counts on the PCGVolume instead: sum(c.get_instance_count() for c in volume.get_components_by_class(unreal.InstancedStaticMeshComponent)).

⚠️ Prerequisites

Both plugins must be enabled in the project's .uproject:

{ "Name": "PCG", "Enabled": true },
{ "Name": "PCGPythonInterop", "Enabled": true }

Verify they're loaded before executing:

import unreal
assert hasattr(unreal, 'PCGGraph'), "PCG plugins not enabled — check .uproject"

Creating a PCG Graph Asset

import unreal

factory = unreal.PCGGraphFactory()
asset_tools = unreal.AssetToolsHelpers.get_asset_tools()
graph = asset_tools.create_asset('MyPCGGraph', '/Game/PCG', unreal.PCGGraph, factory)
unreal.EditorAssetLibrary.save_asset('/Game/PCG/MyPCGGraph')

Loading an Existing PCG Graph

import unreal

graph = unreal.EditorAssetLibrary.load_asset('/Game/PCG/MyPCGGraph')
# graph is a PCGGraph — access nodes, input_node, output_node directly

Or search first with the Asset Registry (or Epic's AssetTools toolset):

ar = unreal.AssetRegistryHelpers.get_asset_registry()
hits = ar.get_assets_by_class(unreal.TopLevelAssetPath("/Script/PCG", "PCGGraph"))

⚠️ PCGGraph.is_standalone_graph is deprecated (issue #465) — it logs a deprecation warning in 5.8. Use the graph's GraphUsageContext / usage flags instead of is_standalone_graph to tell a top-level graph from a subgraph.

Adding Nodes

Use add_node_of_type(SettingsClass) — returns a (PCGNode, PCGSettings) tuple. All Settings classes are direct attributes of the unreal module — no find_class needed:

import unreal

# Surface Sampler node
node, settings = graph.add_node_of_type(unreal.PCGSurfaceSamplerSettings)
settings.set_editor_property('points_per_squared_meter', 5.0)
settings.set_editor_property('point_extents', unreal.Vector(50, 50, 50))
node.set_node_position(200, 0)

# Static Mesh Spawner node
spawner_node, spawner_settings = graph.add_node_of_type(unreal.PCGStaticMeshSpawnerSettings)
spawner_node.set_node_position(500, 0)

Settings can also be read/written after creation via node.get_settings():

settings = node.get_settings()
settings.set_editor_property('unbounded', True)
print(settings.get_editor_property('points_per_squared_meter'))

Identifying Nodes

node.get_editor_property('node_title') is None by default. Identify nodes by their settings class:

for node in graph.nodes:
    print(type(node.get_settings()).__name__)

Note: the nodes array does not include the graph's Input and Output nodes. Access those via graph.get_input_node() and graph.get_output_node().

Discovering Available Node Types

import unreal
node_types = sorted([x for x in dir(unreal) if x.endswith('Settings') and 'PCG' in x])
print(node_types)

Task Index

Task	Sub-doc	Sample script
Create / load a PCG graph	(above)	scripts/build_pcg_graph.txt
Add / identify / discover nodes	(above)	scripts/build_pcg_graph.txt
Connect pins / position / remove	workflows.md	scripts/build_pcg_graph.txt
Place a PCGVolume and generate	workflows.md → Placing a PCGVolume	—
Surface Sampler → Mesh Spawner	workflows.md → Full Example	—
Settings property names / SMS config	reference.md	—

Sub-docs

• workflows.md — pin inspection/connection, node positioning/removal, editor notify, save/delete, placing a PCGVolume + generating, the full surface-sampler→mesh-spawner example, subgraphs.
• reference.md — settings property names, lowercase struct properties, Static Mesh Spawner config, common node settings classes.

Sample scripts (run via execute_python_code)

• scripts/build_pcg_graph.txt — create a PCG graph and add Surface Sampler + Static Mesh Spawner nodes (native PCG API).