---
canonical: https://tanveerriaz.me/blog/digital-twin
format: text/markdown
---

I sent Claude one selfie and one line: create a digital version of me in Three.js, make no mistakes.

It made mistakes. So did I.

[Open the twin full screen ↗](/blog/digital-twin/assets/twin)

## Try one: a cartoon of someone like me

The first version was built by hand from shapes. A khaki cap. Black glasses. About 40,000 beard strands. A laptop with stickers. It typed and it waved.

It was a nice cartoon of a man with a beard. It was not me.

## Try two: my face on the wrong head

So I asked for the same face. Claude measured my face in the selfie with 478 landmark points and wrapped the photo over that shape. I sent two more photos and it added my hairline.

Now the eyes were mine. Everything around them was still made up. The skin tone changed at the cheek. The ears were modelled. One side of my face was a mirror of the other, because the camera only saw one side.

My reply was short: “No, man, this is not me.”

## Try three: five views, one head

The fix was not a better prompt. It was better input. I gave it a reference sheet of my head: front, both three-quarter views, both profiles and a close-up of skin and beard.

That changed the method. The two profiles gave the outline of my forehead, nose, lips, chin and the back of my head. The front view gave the width. Skin, beard, hair and ears were baked from the front and the two profile photos into one texture, on one continuous surface of about 98,000 points.

The glasses came off the photos and went back on in 3D, traced from the outline of my own frames.

The whole thing is one HTML file with Three.js inside it. No server.

## What is still not real

Nobody photographed the top of my head or the centre of the back. That hair is generated in my hair colour. The body is still the simple cartoon one. The face does not blink or smile. The head turns to follow you and the arm waves. That is all it does.

## What I took from it

“Make no mistakes” is not a spec. A selfie is one view of a 3D object. Asking harder does not add the sides the camera never saw.

One photo gives you a mask. Five views give you a head.

Next time I start with the reference, not the adjective.

[Open the twin full screen ↗](/blog/digital-twin/assets/twin)