Return to Answer

This is a workaround and I hope there is (or will eventually be) an easier way to access point cloud data in the API.

In your GN tree add an Instance on Points node, plug in a Mesh Line set to a Count of 1 to get a single vertex at $(0, 0, 0)$.

In theory we should be able to access the vertices' geometry but for now the geometry is not "real". We need to realize the instances with a Realize Instances node.

Then, we need to access the evaluated mesh, after the GN modifier has been evaluated. For that we can use the context dependency graph.

import bpy
import numpy as np

depsgraph = bpy.context.evaluated_depsgraph_get()
obj = bpy.context.active_object.evaluated_get(depsgraph)

coords = np.zeros(len(obj.data.vertices) * 3, dtype=float)
obj.data.vertices.foreach_get("co", coords)

print(coords)  # This is a flattened array

coords = coords.reshape(len(obj.data.vertices), 3)

print(coords)  # This is a 3D array

Result :

This is a workaround and I hope there is (or will eventually be) an easier way to access point cloud data in the API.

In your GN tree add an Instance on Points node, plug in a Mesh Line or a Points node set to a Count of 1 to get a single vertex at $(0, 0, 0)$.

In theory we should be able to access the vertices' geometry but for now the geometry is not "real". We need to realize the instances with a Realize Instances node.

Then, we need to access the evaluated mesh, after the GN modifier has been evaluated. For that we can use the context dependency graph.

import bpy
import numpy as np

depsgraph = bpy.context.evaluated_depsgraph_get()
obj = bpy.context.active_object.evaluated_get(depsgraph)

coords = np.zeros(len(obj.data.vertices) * 3, dtype=float)
obj.data.vertices.foreach_get("co", coords)

print(coords)  # This is a flattened array

coords = coords.reshape(len(obj.data.vertices), 3)

print(coords)  # This is a 3D array

Result :

This is a workaround and I hope there is (or will eventually be) an easier way to access point cloud data in the API.

In your GN tree add an Instance on Points node, plug in a Mesh Line set to a Count of 1 to get a single vertex at $(0, 0, 0)$.

In theory we should be able to access the vertices' geometry but for now the geometry is not "real". We need to realize the instances with a Realize Instances node.

Then, we need to access the evaluated mesh, after the GN modifier has been evaluated. For that we can use the context dependency graph.

import bpy

depsgraph = bpy.context.evaluated_depsgraph_get()
obj = bpy.context.active_object.evaluated_get(depsgraph)

print([v.co for v in obj.data.vertices])

Result :

This is a workaround and I hope there is (or will eventually be) an easier way to access point cloud data in the API.

In your GN tree add an Instance on Points node, plug in a Mesh Line set to a Count of 1 to get a single vertex at $(0, 0, 0)$.

In theory we should be able to access the vertices' geometry but for now the geometry is not "real". We need to realize the instances with a Realize Instances node.

Then, we need to access the evaluated mesh, after the GN modifier has been evaluated. For that we can use the context dependency graph.

import bpy
import numpy as np

depsgraph = bpy.context.evaluated_depsgraph_get()
obj = bpy.context.active_object.evaluated_get(depsgraph)

coords = np.zeros(len(obj.data.vertices) * 3, dtype=float)
obj.data.vertices.foreach_get("co", coords)

print(coords)  # This is a flattened array

coords = coords.reshape(len(obj.data.vertices), 3)

print(coords)  # This is a 3D array

Result :

This is a workaround and I hope there is (or will eventually be) an easier way to access point cloud data in the API.

Instead of instancing points, create a new object, eg. a cube, go into edit mode, select everything and right click > Merge Vertices > At Center. That way you have an object with a single vertex at world origin.

In your GN tree add a Point Instance node and choose your single vertex object.

In theory we should be able to access the vertices' geometry but for now the geometry is not "real". We need to trick the modifier into creating real geometry. There are multiple nodes for which it will work. I plugged a default and blank Attribute Math node before the modifier output.

Then, we need to access the evaluated mesh, after the GN modifier has been applied. For that we can use the context dependency graph.

import bpy

depsgraph = bpy.context.evaluated_depsgraph_get()
obj = bpy.context.active_object.evaluated_get(depsgraph)

print([v.co for v in obj.data.vertices])

Result :

This is a workaround and I hope there is (or will eventually be) an easier way to access point cloud data in the API.

In your GN tree add an Instance on Points node, plug in a Mesh Line set to a Count of 1 to get a single vertex at $(0, 0, 0)$.

In theory we should be able to access the vertices' geometry but for now the geometry is not "real". We need to realize the instances with a Realize Instances node.

Then, we need to access the evaluated mesh, after the GN modifier has been evaluated. For that we can use the context dependency graph.

import bpy

depsgraph = bpy.context.evaluated_depsgraph_get()
obj = bpy.context.active_object.evaluated_get(depsgraph)

print([v.co for v in obj.data.vertices])

Result :