Alt-comic redraw

This one started as somebody else’s work. The upstream is an open-source skill for redrawing photos as flat alt-comics — the kind of quietly absurd, emotionally flat illustration that reads like a single panel torn out of an indie comic where nothing much happens.

What we changed was mostly the parts that break. The original was an agent workflow: four steps of looking at the image, passing file paths around, and handing results back. None of that survives here, because our pipeline hands the model your photo directly — the workflow scaffolding was solving a problem we do not have.

The parts we rewrote were about text. The original asked for a handwritten line but said nothing about spelling, and image models are famously bad at lettering — a smudged or misspelled caption ruins an otherwise clean panel, and there is no way to fix it after the fact. So the text section became an explicit instruction to read the photo first and write about something visible in it, with a hard no-typos rule attached. We also found three leftovers from whatever the original author had actually been making — a Mother’s Day theme, a greeting-card phrase, and an oddly specific ban on coffee references. The first two we folded into a general cliché ban. The coffee one we kept, on the theory that it was probably load-bearing.

The last change was to quantify what “leave lots of empty space” means. Vague amounts get ignored; 30–50% of the canvas does not. That single substitution did more for the finished look than anything else we touched.

It went out after one round of testing across eight photos — portraits, food, a street cat, a lake, a cathedral. Flat styles like this one are unusually well-behaved: because the instruction is to remove detail rather than add it, there is much less room for the model to invent something that was never there.