Skip to content
← The Workbook

No. 36 · 2026Concept

Carface

The back of a car has a face. Headlights and taillights are eyes. The bumper is the chin. The license plate is the mouth, sometimes the teeth. This is a tool that reads the rear-end industrial design of any car you photograph and translates the personality into a human portrait. Angry taillights stay angry. The cute aloof one stays cute.

Status
Concept
Year
2026
Stack
Computer Vision · Image-to-Image · AI · Industrial Design · Portraiture

The Idea

The back of cars have faces. Some are upset. Some are aloof. Some are dopey. Some are cocky. The taillights set the eyes, the bumper sets the chin, the trunk lid sets the brow. License plate is sometimes a mouth, sometimes a chin, sometimes nothing at all. The personality is right there in the industrial design and people read it without realizing they're reading it.

This tool reads it on purpose, then translates the personality to a human portrait that retains the same expression, mood, and energy as the original car.

The Design Vocabulary

A few patterns the model would learn first:

  • Taillight angle = mood. Slanted-down outer edges read angry or annoyed. Slanted-up read surprised or eager. Round and centered read aloof or unbothered.
  • Bumper slope = expression. Down-curving lower bumpers read grumpy. Up-swooping ones read smug or smiling. Flat ones read deadpan.
  • Trunk-lid height = brow. Low trunk creates a heavy brow line, the angry stare. High trunk lifts the eyebrows, an open look.
  • Exhaust placement = stance. Centered single = serious. Twin pipes flared out = aggressive. Quad = trying very hard. Hidden = polite.
  • License plate = mouth, but only when it adds to the read. Sometimes it's buck teeth. Sometimes it's a closed mouth. Sometimes it's nothing - the bumper does the mouth work and the plate just sits.

The system picks per car, doesn't force the same anatomy onto every photo.

What the Tool Does

Input: a photo of the back of a car.

Output: a portrait of a person whose face carries the same personality the car's rear is broadcasting. Same mood, same posture of features, same energy. Not a literal feature-for-feature translation. A character translation.

The car in the photo informs the portrait directly: same lighting, same angle, same color temperature, same era of style. A 1972 Mercedes 280 reads different from a 2024 Hyundai Ioniq, and the portraits should land in different worlds. The output is a faithful interpretation of THIS car, not a generic "person face."

Pairs With Plate Karma

Same dashboard hardware path: Jetson plus camera plus button. The bad-driver-database project (Plate Karma) is the practical one. Carface is the same camera looking at the same car, drawing different conclusions. The cut-me-off jerk you logged in Plate Karma also gets a portrait that captures, with great fidelity, the personality of his Ram 1500.

Status

Concept. The interesting work is the design-vocabulary dataset (the mapping of car-rear features to human-face moods) and the prompt engineering for the image-to-image step. Front-end is trivial. Side-project tier.