Build write-up · in progress · 8 min read

How I’m building Faunaform

Seven Minnesota animals, each drawn for one hour of one day, to be screen printed by hand in four flat inks. An image model draws, software I wrote with Claude Code makes every drawing printable, and every call about the art is mine. Nothing is printed yet; the first proof is next.

Role
Everything: art direction, brand, the studio’s software, the shop’s design
Built with
ChatGPT to draw, Claude Code for the software, Vectorizer.AI for curves
Runs as
A private studio site and a print pipeline: drawings in, film-ready screens out
Status
In progress: supplies ordered, the first proof next
Faunaform Faunaform

Seven Minnesota animals, each at its own hour of the day.

Seven Minnesota animals, each at its own hour of the day.

My approach

  1. 01Research

    • What the animal really does here
    • Old prints and new work, one idea each
    • Field marks written into the brief
    One print, start to finish

  2. 02Draw

    • An image model draws from a brief
    • Three takes, one pushed further
    • Rounds until nothing’s left to change
    Who does what

  3. 03Make it printable

    • Every pixel one of four inks
    • Shapes too thin to print, marked
    • Smooth curves, then four screens
    Making it printable

  4. 04Print

    • The bear first, on paper and birch
    • Buyer research before a price
    • Fairs asked about AI first
    How I’ll check it

The problem

An image model can draw a striking wildlife print in a minute. It can’t tell you whether the print will print, or whether the bear is a bear.

Software an agent helped build has the same problem: someone has to answer for every call.

The collection is one idea, One day of light: seven animals, each at its own hour of one day, from the owl before dawn to the hare at dusk.

  1. The great horned owl before dawn: perched on a bare branch over snow, a pale moon setting into a pink band at the horizon. Before dawn Great horned owl
  2. The sandhill crane at sunrise: flying low over cattails, a red-orange sun on a peach horizon and its light on the water. Sunrise Sandhill crane
  3. The moose in morning mist: drinking in a pond under pale fog, spruce on the far shore and a gold sun rising. Morning mist Moose
  4. The common loon at noon: on a deep blue lake with two chicks behind her, water lilies in front, pines on the far shore and a small sun high in the corner. Noon Common loon
  5. The black bear at golden hour: walking through a berry patch between birch trunks, an orange sun low behind them. Golden hour Black bear
  6. The red fox at sunset: sitting in tall gold grass, a red sun on the horizon and a dark tree behind. Sunset Red fox
  7. The snowshoe hare at dusk: white on snow between spruce heavy with it, a warm moon rising over a violet band. Dusk Snowshoe hare
The seven, simulated at six inches; nothing is printed yet. Under each, its four inks in the order they’re pulled.

Who does what

Every step of the work says who does it, which answers the question everyone asks about AI and art.

  • Me. The idea, the research, every choice and the final say. And the printing.
  • An AI image model, ChatGPT. It draws from a written brief, about a minute a take.
  • An AI assistant, Claude. It helps with research, turns my direction into briefs and writes the studio’s software.
  • Studio software. Ordinary code, not AI: it fixes the colors, measures the shapes and makes the screens, the same way every time.
39
scripts in the studio’s software, written with Claude Code in the first week
81
briefs, each a file, so any drawing can be asked for again
19
rules in the print standard every drawing is checked against

Every drawing also records its round, its take and its source, a habit from Lucy Learns. That’s the answer when a fair asks how a piece was made.

One print, start to finish

The black bear at golden hour, drawn from 16 to 20 October 2026.

Every print starts from something the animal really does here. Black bears feed by day and blueberries ripen in August, so the bear is in the berry patch, head down.

Old prints, one idea each

Claude searched three museums’ open collections for old prints that solved a problem like this one. Each went on the board for one idea: a pose, a shape, a way to place the sun.

A woodblock page of two bears, each one dark shape on short legs with nothing drawn inside it. One dark shape
One dark, round mass filling the sheet over pale grasses, with the moon tucked behind it. One round mass
Sparrows in snow among nandina, its red berries drawn as plain round dots in loose clusters. Berries as dots
A dark paper stencil cut with an even carpet of mallow leaves, some solid and some open. An even carpet of leaves
Kuwagata Keisai, Abbreviated Drawing Styles for Birds and Animals, 1797, The Metropolitan Museum of Art. Kamisaka Sekka, Autumn Night, 1909–10, Cleveland Museum of Art. Utagawa Hiroshige, Sparrows Flitting about Snow-covered Nandina, about 1840, and Stencil with Design of Mallows, Japan, late 19th century, both Art Institute of Chicago. All public domain. The dark shape and the dotted berries made it to the print; the big sun didn’t.

New work, on Cosmos

For contemporary work Claude searched Cosmos, a site for collecting images, the same way. Neither set goes to the image model; their ideas go into the brief, in words.

A poster in flat orange, teal and cream: a grizzly at the center, ringed by the park’s other animals in circles. Few flat inks
A matchbox label printed in brown and green on tan: a bear sitting among ferns and trees, the word Medve under it. Two inks
A polar bear as one white shape on black, with only its eyes and nose drawn in. One shape
A bear’s head as one round blue shape cut in rings, with a pale muzzle. Cut in one ink
From Cosmos: Tom Whalen, The Iconic Wildlife of Glacier National Park, for The 59 Parks Print Series. Medve, a matchbox label from Eastern Europe, its maker not named. A polar bear and a bear’s head in blue, their artists not credited on Cosmos.

Getting the animal right

A black bear isn’t a small grizzly, so the brief spells out the marks: a straight face, no shoulder hump, tall round-tipped ears, a tan muzzle.

The brief, then the drawing

The bear’s brief opened: “An American black bear on all fours in a paper birch grove in August, side-on facing right, head lowered toward a patch of lowbush blueberries.”

I ask for three takes of a new animal and push one further. The bear began as a dark dome under a huge sun, planned in wood; a day later it stood among birches at golden hour.

The first concept: a dark bear made of wood-grain pieces, head down at red berries, under a huge yellow sun on a pale wood ground. 16 October · planned in wood
Golden hour: the bear in aubergine among birch trunks and blue and amber plants, a low orange sun behind it. 17 October · golden hour

Saying what’s wrong

Most of the work is saying exactly what’s wrong. The first square bear had a hazy sky and a row of tiny triangle trees. Every ink is flat, so it went back, and take two corrected both.

Take one: the square bear among birches and berries, its sky softly blurred and small triangle trees along the far shore. Take one
Take two: the same composition with a flat sky in hard bands and the triangle trees gone. Take two, kept

Looking hard still misses things. Count the bear’s legs in both takes: there are five. A dark shape under its rump reads as a fifth leg.

I approved the bear with all five on 19 October. It was caught the next day, and the legs were redrawn from a public-domain photo of a black bear walking.

Making it printable

Here the software takes over, because this part has to come out the same every time.

Exact inks

The image model’s colors are close, never exact. The software sets every pixel to one of the four inks or the bare paper, so every shape is one flat color with a clean edge.

The bear simulated in four inks on white paper: amber light, orange sun, grayed blue sky and shadow, and an aubergine bear. Four inks and the paper
A close-up of the sun between two birch trunks, every shape one flat ink with a clean edge. Up close
Simulations of the print, not photographs.

Will it print?

At six inches, a shape thinner than about 1.5 mm won’t hold ink. The software marks each one in magenta, and the drawing goes back with one rule: widen, never remove.

The bear in gray with every place too thin to print at six inches marked in magenta, mostly in the plants and the birch marks. Too thin, marked
The final bear in full color, walking on four legs with short bear paws, its plants and birch marks widened. Final, 20 October
4.0 → 2.4%
of the bear too thin to print, before and after its redraw
7.6 → 3.3%
of the fox
4.2 → 2.0%
of the loon

Smooth curves, then four screens

Up close, pixels turn to stairs, so Vectorizer.AI redraws every edge as a smooth curve. Then the software splits the print into its screens, with registration marks on every film.

The bear’s eye enlarged, its edges stepped in visible pixels. Pixels
The same eye redrawn in smooth curves. Curves
The light screen alone: every sunlit plane in amber. Light
The sky screen alone: the sky and every shadow in grayed blue. Sky
The sun screen alone: the low sun and a few berries in orange, almost all bare paper. Sun
The deep screen alone: the bear and the dark marks on the birch bark in aubergine. Deep
The eye in pixels and in curves, then the bear’s four screens, printed from light to deep.

The model widens the search. Code holds the production truth.

Three decisions that shaped it

The medium came last

It began on 14 October as laser-cut wood, then veneer, matboard and collage, each judged on simulated renders. By the 16th it was screen printing. Four media in three days cost nothing but compute.

The fox at sunset on smooth white paper, a white margin round the print. Smooth paper
The same print on textured cotton paper. Textured cotton
The same print on a Baltic birch panel, the bare wood showing wherever no ink lands. Baltic birch
The fox on the first run’s three grounds. Simulations, not photographs.

One idea, held across seven pieces

Every print takes its four inks from its own hour’s sky: gold where the sun lands, blue-violet in the shadows. Every round had to keep that idea or go, and more than ten went.

Say what AI did, in plain words

“AI-assisted,” never “AI-generated,” and no render shown as finished. A label that says AI can change what people will pay, so the wording has to be honest and careful.

How I’ll check it

  • The first proof. The bear, on paper and then on birch, with every rule confirmed or rewritten against it.
  • Buyers before a price. Two rounds of six to ten people.
  • Fairs asked first. Some here exclude AI-assisted work.

What it has to do with product design

It’s the argument of this site, applied to art: a standard every piece is checked against, a decision log, an image model to widen the search, software to hold what must not drift, and a person deciding what ships.

That’s the infrastructure behind great design, at the scale of one studio and one bear.