Prop to Image

A workflow for reimagining physical props through images and preparing selected designs for 3D reconstruction.

Creative toolsAI3D
A 3D model of a dark blue-black film prop with brass trim and wood grips, rendered in Blender
The selected concept reconstructed in 3D with TRELLIS.2, after a material pass in Blender. A Blender render.
Status
Local tool, one prop taken through the chain
Tools
Python CLI, Codex image edit, TRELLIS.2, Blender
Links
Source

Start from a photo of a real object, explore new exteriors with an image model, pick one, and hand the picked image to image-to-3D reconstruction. It also runs the other way, from a generated idea to a model split for printing. The tool keeps the source, each request, each candidate and the pick linked, so a 3D model can be traced back to where it came from.

The reconstructed model turning in its local Three.js viewer.

Decisions

Keep the source and every candidate

The CLI preserves the source photo, prepares each image request, registers results and builds a comparison gallery. A failed or blocked generation is recorded too, so nothing disappears between rounds.

Picking is a recorded step

A person records the pick, with a checksum of the exact image, before it goes to reconstruction. Choosing a look stays separate from approving it for the film.

Three images in sequence: an orange and white toy on a wooden floor, a generated dark prop concept with brass and wood, and the 3D model of that concept
Diagram of one prop's path from a real object, left to right: the original toy photo, the generated concept that was picked, and the reconstructed 3D model.
Three images in sequence: a generated concept of a film prop on a workbench, its 3D reconstruction, and the model split into eight flat-backed print pieces
Diagram of the other direction, from an idea to a print, left to right: a generated concept, its TRELLIS.2 model, and the model split into eight flat-backed pieces for a support-free print. Not yet printed.