Point clouds¶

pl.add_mesh(cloud, style="points") and style="points_gaussian" render each vertex of the actor's dataset as a discrete particle instead of triangulating it into a surface. The bridge dispatches both to a native bpy.types.PointCloud data-block rather than building a sphere mesh per point. Cycles renders point clouds as per-point spheres directly, so memory + render cost scale with N_points instead of N_points * N_geom_verts.

Opaque points (style="points")¶

import numpy as np
import pyvista as pv

rng = np.random.default_rng(seed=42)
points = rng.normal(0.0, 0.8, size=(2_000, 3)).astype(np.float32)
cloud = pv.PolyData(points)
cloud["height"] = points[:, 2]

pl = pv.Plotter(off_screen=True, window_size=(800, 600))
pl.add_mesh(
    cloud,
    style="points",
    scalars="height",
    cmap="viridis",
    point_size=8,
    show_scalar_bar=False,
)
pl.camera_position = [(3.0, 3.0, 2.0), (0.0, 0.0, 0.0), (0.0, 0.0, 1.0)]
pl.set_background("#0c1018")

pl.blender.render("points.png", samples=32)
pl.close()

The bridge builds a Principled BSDF coloured either by the active scalar field (scalars= was set) or by prop.color (no scalars). Per-point colours land on a "scalars" POINT-domain Color attribute on the point cloud, so the material's ShaderNodeAttribute reads them exactly like the mesh path does.

Gaussian splats (style="points_gaussian")¶

pl = pv.Plotter(off_screen=True, window_size=(800, 600))
pl.add_mesh(
    cloud,
    style="points_gaussian",
    scalars="height",
    cmap="inferno",
    point_size=20,
    show_scalar_bar=False,
)
pl.camera_position = [(3.0, 3.0, 2.0), (0.0, 0.0, 0.0), (0.0, 0.0, 1.0)]
pl.set_background("#000000")

pl.blender.render("gaussian.png", samples=32)
pl.close()

For style="points_gaussian" the bridge swaps the Principled BSDF for a Transparent + Emission mix driven by a camera-facing falloff: alpha = max(0, N . V) ** 4, where N is the per-point sphere's surface normal and V is the view direction. The result is bright at the centre of each splat (where the sphere faces the camera) and fades smoothly to transparent at the silhouette — the conventional soft-splat look used by particle systems and Gaussian-Splatting datasets, without a per-point centre accessor or geometry-nodes instancing.

How the bridge picks the path¶

The decision happens in scene.py:_detect_point_cloud_mode, which inspects the actor's mapper class first (pv.PointGaussianMapper) and falls back to prop.style == "Points". The mesh path stays unchanged for surface / wireframe rendering.

Constraints and caveats¶

• Point size is in pixels (PyVista) vs. world units (Cycles). PyVista's point_size is screen pixels; Cycles' PointCloud uses world-space radii. The bridge converts via POINT_SIZE_TO_WORLD_RADIUS = 0.005, tuned for unit-bound data. For very small or very large coordinate extents, override by mutating the radius attribute on the returned PointCloud or by scaling the actor before pl.add_mesh.
• Point-data scalars only. PointCloud has no "cell" concept, so cell-data scalars on a points actor are not surfaced. Convert with mesh = mesh.cell_data_to_point_data() first.
• No interactive viewport handling yet. pl.blender.show() doesn't draw the cloud in the VTK overlay; the Cycles overlay does render it correctly. Drag preview falls back to whatever VTK chooses to draw for point styles.