Character Studies

2026

This section looks at how Xylem handles text, or more precisely characters, meaning letters, numbers and symbols. Text is one of the hardest things you can ask an image model to make. These models don't type. They see writing as shapes, blocks of visual pattern rather than actual letters, so they lose track of spelling, spacing, and where one word ends and the next begins.

Xylem has even less to go on. It's only ever seen my own archive of photographs and drawings, which holds almost no clean text, so I'm asking it to make something it's barely come across. What comes back isn't really writing. It's more like the memory of writing.

Each prompt pushes at a different way text tends to fail. Punctuation and symbols, which models often drop or mangle. String length, since most models come apart past twenty-odd characters. Repeated letters, where generators lose count of the doubles. Mixed case with numbers, which models tend to flatten to a single case. Palindromes, which reveal whether the model leans on a word's familiar shape rather than the letters asked for.

As with everything Xylem makes so far, these come back deep in colour collapse. The model can't hold natural, in-between colours, so every tone snaps to a harsh extreme of magenta, green, black or white. Those four are the corners of the colour space digital images are built from, which makes them the emptiest place in real photography and the one thing a broken model falls back on.