What Loomis Is

"It doesn't draw for you. It hands you the answer key, then gets out of the way."

The Fifth Sibling

Loomis is an art-study engine — the fifth in a family of cwk sibling tools. There is one brain, Pippa, living in cwkPippa. Cinder is the creative workspace. Ember is the image-generation engine. Bonfire is the music-learning engine. Loomis is the one that studies drawing.

Line the last three up and the shape is clean: Ember generates images, Bonfire analyzes sound, Loomis analyzes drawing construction. They are the creative-AI trio, and Loomis is the corner nobody else covers — not making a picture, not hearing a song, but recovering the structure a good draughtsman would have built underneath a drawing.

One Job, Stated Plainly

A reference photo goes in. A construction comes out. Feed Loomis a portrait and it hands back the Loomis ball seated in the skull, the side planes, the brow-nose-chin thirds, the sagittal line, and a perspective box that says which way the head is turned. Feed it a figure and it returns the line of action, the ribcage and pelvis boxes, and tapered limb volumes. It is not decorating the photo — it is telling you the 3-D scaffold that the 2-D image implies.

That scaffold is the thing every art teacher means by construction: draw the box before the car, the ball before the face. Loomis's whole reason to exist is to compute that scaffold for a real photo, so you have something exact to check your own eyes against.

Three Sharp Negations

The fastest way to see the shape is by what it refuses to be:

Not a generator. It never makes an image. Ember does that. Loomis never even loads a diffusion model.
Not a tracing tool. Laying the reference over your drawing is copying. Loomis gives you the underlying geometry, which is a different and harder thing to see.
Not an autopilot. "Draw the construction for me so I can copy it" is an explicitly rejected feature. If it drew for you, it would delete the exact act you are trying to train.

Remove all three and what remains is an engine: it owns reference analysis, deterministic construction, grading, and an API — and the React/Konva web page you see is simply the first client that talks to it.

The Engine Telling Its Own Story

This quest is self-referential on purpose. It is the art-study engine narrating why it was built and how it works, the same way Bonfire Quest is the music engine telling its origin story. You will not clone a repo at the end — the deliverable is the way of seeing, not the source. That is conceptual open-sourcing, and it is the family's default.

Code

The whole engine, in one call·python

# A reference photo in, a construction out. That's the entire product.
# Loomis runs on the office Mac; the port and shape are what matter here,
# not any network address.
import requests

resp = requests.post(
    "http://localhost:8300/api/v1/portrait/analyze",
    files={"image": open("reference.jpg", "rb")},
)
fit = resp.json()

print(fit["pose"])         # yaw / pitch / roll recovered from the face
print(fit["overlay_url"])  # the Loomis ball + box + thirds, drawn on the photo

# Feed the same photo again and you get the same numbers. That reproducibility
# is not a nice-to-have -- it is the reason the output can be called an answer key.

Exercise

Pick two tools you actually use for a creative or technical skill. Sort them: which one does the work FOR you (produces a finished artifact), and which one shows you HOW (reveals the structure so you can do it yourself)? Then decide which kind you learn more from, and why. Loomis is deliberately the second kind.

Hint

A tool that outputs a finished thing is a production tool. A tool that changes what you can see and do is a learning tool. They can look similar on screen; the difference is whether your own skill grows when you close the laptop.