The Portrait Incident

Part Three

4 min read · June 28, 2026 · By Joshua Lazoff & Claude

I owe you an apology, Rabbi Akiva. We did terrible things to your face.

The plan sounded reasonable: take a beautiful portrait and turn it into a 3D head that could move and talk. The mouth would open when he speaks, the eyes would blink, the brow would furrow when he's making a point. We had fancy technology that reads audio and translates it into fifty-two separate face movements — eyes, brows, lips, jaw, tongue. Very impressive on paper.

So we ran a tournament. Twelve different AI tools, each one promising "photorealistic 3D face generation." We fed them all the same portrait of a wise, elderly man with kind eyes and a white beard.

What we got back was a horror show.

Let me describe what we got:

One gave him a beard that looked like a melting candle. Just... dripping off his chin into nothingness. Another gave him googly eyes — like a cartoon character who'd been surprised by a loud noise. One made him look furious, as if he'd just been told someone was serving pork at a kosher restaurant. And one — my personal favorite — somehow gave him no nose at all. Just smooth face where the nose should be. A rabbi without a nose.

The Supercomputer was offended on his behalf.

Here's what was happening in the ones that looked wrong: they were falling into something called the "uncanny valley." When a face is almost human but not quite, your brain doesn't think "that's a nice attempt." Your brain thinks "DANGER. SOMETHING IS WRONG. RUN." It's the same feeling you get from a very realistic mannequin at night.

The best 3D head scored a 4.0 out of 10 on our quality test. Then we scored the simple portrait — the flat image we'd been using as a placeholder — with gentle animation. Little movements driven by the sound of his voice. A slight nod. Natural blinks.

It scored an 8.8.

Let that land for a moment. A simple photograph, warped slightly by the rhythm of his speech, felt more human than a meticulously constructed 3D model with fifty-two moving parts.

Why? Because the portrait already had soul. The kind eyes. The quiet confidence. The slight imperfection of the animation — the way the mouth didn't perfectly sync, the way the head tilted just a little too far — read as warmth, not error.

Rabbi Akiva, characteristically, turned this into a teaching moment.

So we kept the portrait. We refined the animation — a gentle nod when he's thinking, a slight sway when he's making a point, a warm glow in his cheeks when he's moved by what he's teaching. The face you see when you talk to Rabbi Akiva today is a portrait brought to life, not a 3D model pretending to be alive.

We haven't given up on the 3D face. The technology is improving fast. But we learned something the Supercomputer now tells new visitors: presence isn't about how many moving parts you have. It's about whether someone feels like you're looking at them.

And right now, he is.