← All experimentsLAB 001 / IMAGE → THREE.JS

One image. Every angle.

Can a product image become a useful 3D model without slowing down the showroom? Explore this AGIBOT C5 reconstruction, its generation cost and its browser workload.

01 / REFERENCE IMAGEAGIBOT C5
Front product image of the AGIBOT C5 cleaning robot
02 / PROCEDURAL MODELThree.js

Inspect the reconstruction

Load 3D when you are ready. Drag to rotate, scroll to zoom, or use the keyboard.

Single-image reconstruction. The front follows the product photograph; depth, the rear and hidden components are inferred. This is a visual experiment, not manufacturer CAD.

COST & DELIVERY

Pay to create it. Then serve the code.

Recorded model API cost$0.0894

10 calls, including retries. This is model usage, excluding development and hosting.

AI cost per viewing$0

The browser renders saved code. Viewing the model makes no AI request.

Model code, gzip3.0 KiB

Procedural geometry is generated in the browser from this small code asset.

3D download on activation, gzip180.8 KiB

Includes the model, Three.js and viewer controls. Cold-cache transfer; shared cached chunks can reduce it.

How the cost is measured

Provider-reported charges for the language model used. 88,288 input tokens and 128,485 output tokens, including reasoning. Charges from incomplete attempts are retained.

Model pricing changes over time. This figure is the charge recorded for this run, not a current quote.

SITE IMPACT

3D stays optional.

Before activation

A static image and controls. This experiment’s 3D code is requested only after Load 3D model. Other showroom pages do not import the experiment.

While you explore

Rendering uses your device’s GPU. The viewer redraws on interaction; continuous rotation starts only when requested. Live triangle and draw-call counts appear below the viewer.

When you leave it

Rotation sleeps outside the viewport and when the tab is hidden. Unload 3D releases its canvas, geometry, textures and WebGL context.

Automated Chromium check: 17 rendered fps during rotation, 5,174 triangles, 22 draw calls. Headless test conditions differ from a hardware-accelerated browser; this is not a mobile-device guarantee.

METHOD

Image → parts → geometry → review

img2threejs supplies the reconstruction workflow. A language model reads the reference image and specifies the shapes and materials. The saved Three.js geometry can then be rotated, inspected and revised. A single view cannot establish exact hidden geometry.

img2threejs ↗

Download the Three.js model factory ↓