Exporting a notebook to HTML with working pyvista scenes

A Jupyter notebook full of pyvista scenes can be converted to a single HTML file in which every scene stays fully interactive — rotate, zoom, pan — in any browser, with no Python behind it. The catch is the Jupyter backend.

The trap, and the backend that avoids it

PyVista’s default notebook backend ('trame') is a live widget: the scene in the cell is served by the running kernel, so jupyter nbconvert captures a dead placeholder. This is what makes the export look impossible.

The 'html' backend is the one built for this: it serializes each scene’s geometry into a self-contained vtk.js viewer inside the cell output, so the scene needs no kernel ever again. Available in pyvista ≥ 0.44 (verified here on 0.47.3).

Recipe

import pyvista as pv
pv.set_jupyter_backend('html')   # once, at the top of the notebook

Show plotters as usual, run the notebook top to bottom, then:

jupyter nbconvert --to html notebook.ipynb

To embed only selected scenes, set it per call instead: pl.show(jupyter_backend='html'). For a standalone file per scene, outside any notebook, there is plotter.export_html("scene.html").

Caveats

“Interactive” means the camera. Anything backed by Python — slider widgets, picking callbacks, live updates — dies with the kernel by definition. Surface meshes are the well-trodden path in vtk.js; volume rendering support is partial.
File size is the real constraint. The geometry is embedded in the HTML, so a notebook of block-model contours and topographies can produce a file of hundreds of MB that browsers choke on. simplify(max_error) exists for exactly this:
```
shell = bm.get_contour("assay/Ag/prediction", cutoff,
                       close="above", simplify=1.0)
below = topo.simplify(1.0).clip_meshes([shell])
```
keeps each scene a few MB at sub-metre fidelity (the argument is the largest distance the simplified surface may move, in coordinate units).
The export itself renders. Run the conversion from an environment where pyvista renders off-screen — under WSL that means the conda env with the libstdc++ fix installed (see wsl-pyvista-rendering.md).

One report, no server

plots.Dashboard already writes self-contained HTML for the plotly figures, linked selection included. The 'html' backend is the pyvista counterpart: a complete deliverable — dashboard plus interactive 3D scenes — can ship as plain HTML files that open anywhere.