UI UX

Shilpa3D — Letting People Explore a Sculpture They Can't Physically Reach

Shilpa3D — Letting People Explore a Sculpture They Can't Physically Reach

Shilpa3D — Letting People Explore a Sculpture They Can't Physically Reach

Turns a single 2D photo of a sculpture into an explorable 3D model — built for researchers and viewers who can't physically reach fragile or inaccessible pieces.

View on Figma

View on Figma

View Live

View Live

Duration

Feb-Jul 2025

Organization


GCES

Created on

Updated on

The problem: Many of Nepal's temples and sculptures are fragile, damaged, or hard to access. Shilpa3D reconstructs a 3D model from a single photo for researchers and viewers who can't travel to see the original.

The hardest design problem: A single photo can't fully capture a sculpture's true form — the real question was how to show an uncertain reconstruction honestly, without hiding the uncertainty or making the tool feel unreliable.

The solution: Opacity and thickness controls let people probe the model themselves rather than take a fixed result at face value — the same trust logic as your other work: show the limits instead of asserting false confidence.

Where it stands: Interface fully live (upload, processing, viewer, history); reconstruction model still training, so results are early.

The problem: Many of Nepal's temples and sculptures are fragile, damaged, or hard to access. Shilpa3D reconstructs a 3D model from a single photo for researchers and viewers who can't travel to see the original.

The hardest design problem: A single photo can't fully capture a sculpture's true form — the real question was how to show an uncertain reconstruction honestly, without hiding the uncertainty or making the tool feel unreliable.

The solution: Opacity and thickness controls let people probe the model themselves rather than take a fixed result at face value — the same trust logic as your other work: show the limits instead of asserting false confidence.

Where it stands: Interface fully live (upload, processing, viewer, history); reconstruction model still training, so results are early.

The problem: Many of Nepal's temples and sculptures are fragile, damaged, or hard to access. Shilpa3D reconstructs a 3D model from a single photo for researchers and viewers who can't travel to see the original.

The hardest design problem: A single photo can't fully capture a sculpture's true form — the real question was how to show an uncertain reconstruction honestly, without hiding the uncertainty or making the tool feel unreliable.

The solution: Opacity and thickness controls let people probe the model themselves rather than take a fixed result at face value — the same trust logic as your other work: show the limits instead of asserting false confidence.

Where it stands: Interface fully live (upload, processing, viewer, history); reconstruction model still training, so results are early.

Challenge

Challenge: Nepal has a large number of temples and sculptures, many of them fragile, damaged, or hard to physically access. Shilpa3D takes a single 2D photo of a sculpture and reconstructs it as an explorable 3D model — built for researchers who need to study a piece they can't travel to, and for anyone who wants to see a sculpture's full form without putting the original at risk.

Solution

designed the full user experience — selecting and uploading an image, the processing state while reconstruction happens, how results are presented, downloading the output, and a history of past reconstructions. I also built the frontend in React and Three.js to implement the product.


The real design problem wasn't displaying a finished 3D model — it was displaying an uncertain one honestly. A single photo can't fully capture a sculpture's true form; the reconstruction has to guess at angles and details the camera never saw. I added opacity and thickness controls so people can visually adjust how the model reads — a way to probe the reconstruction rather than take it at face value. It's the same underlying idea as designing trust into a system that can't promise perfection: give the person a way to see the limits themselves, instead of asserting confidence the system doesn't actually have.

Impact

The interface — upload, processing, viewer, history — is built and live. The reconstruction model itself is still being trained, so current results are early and will keep improving. I also designed the project's brand mark.