Work in progress
Pieralongia / Seceda
I sat here. Then I tried to rebuild it.
01 / OVERVIEW
The idea behind it.
I visited Pieralongia near Seceda and sat on top of one of the spires. Afterward, I wanted a good 3D model of the area so I could show people what it was like. The trip photos and video gave me a starting point. I first tried building an interactive scene with Claude, then switched to Codex and continued with editable Blender models. The project is still unfinished: I’m improving the model and working toward a cleaner final product.
02 / THE APPROACH
The visit, the first scene, and the work since.
First, describe the rocks
My first attempt with Claude became an interactive Three.js scene with modeled towers, five viewpoints, and a small seated figure. Asking for a geometry description before building gave me something I could check against the photos. “The big tower” was clearer than “left” or “right,” which changed with the viewpoint. This was a photo-based interpretation, not a photogrammetric mesh.
Continue in Codex and Blender
After switching to Codex, I developed the scene from a simple blockout into a more detailed photo-based model. Python scripts made the geometry editable and the scene reproducible. The early versions still relied on assumptions about the rocks’ heights, spacing, and surroundings.
Give the terrain a measured foundation
A public 2.5 m digital surface model from the Province of Bolzano anchored the terrain. That grid also revealed a limit: it could not describe the vertical walls, overhangs, and fractures that make the spires recognizable.
Keep measured and interpreted geometry traceable
I preserved the native measured grid in one Blender scene. A separate presentation scene adds photo-assisted rock meshes and records the ground edits beneath them. Registered aerial imagery supplies the terrain color; the rock faces remain approximations informed by photographs.
Leave an unsuccessful experiment out
A later TRELLIS.2 image-to-3D preview broke the landscape into thin, fragmented surfaces. No mesh from that test was integrated. It remains a separate experiment within the Codex phase.
Refine the shapes, then the camera
I worked on asymmetric fins, broken shoulders, pale stone on steep faces, grass on the sloping backs, and the gap between the pair. Camera studies mattered too: some views hid the lower rocks behind foreground terrain, while another made their relationship easier to read.
Built with
- Blender
- Python
- Three.js
- Terrain data
03 / A CLOSER LOOK
Watch the builder’s story.
The film follows the work so far using trip references, the supplied Claude scene recording, and saved Blender models and renders. Measured terrain and interpreted rock geometry are identified throughout. The Claude footage includes a later v1.1 display correction made while producing the earlier documentary.
Read the video transcriptDownload the film
Terrain imagery: Orthophoto 2023 © Autonomous Province of Bolzano–South Tyrol, CC BY 4.0. Cropped, stitched, JPEG-reencoded, draped onto terrain, and rendered in Blender for this model.
04 / REFLECTION
What I’m taking forward.
Describing the rocks before building made assumptions visible. Separating measured terrain from interpreted geometry made those assumptions easier to trace. Today I have an editable Blender reconstruction I can explore and use to show the place, but the rocks are still stylized and the cliff faces are not surveyed. I’m still interested in improving the 3D model and making a cleaner final product. The goal is the same one I started with after the trip: help other people understand what it was like to be there.
