tests of creativity

21 Jun, 2026

At the end of the experiment were two quite chastened machine minds. ::chuckle::

To be sure, I didn’t have high expectations for either ChatGPT or Claude, as imagegen isn’t their forté after all. Still, I believed they were capable of something… visually interesting.

I prompted ChatGPT (5.5 Thinking) first:

Here's another 'creativity'-type prompt, and a cold-contextual test.
Can you draw me a scene from Ryokan's poem "A Solitary Mirror," in the style of classical Japanese landscape ink painting (sumi-e)? Preferably vertical in orientation.

It thought and thought, and drew and drew... and just froze up after several minutes. I tried again; and again. Same result. Its excuses:

So next, I prompted Claude (Opus 4.8 High). While it didn't freeze up like its other sibling across the hut, it... used SVG. I suppose sumi-e is tractable with respect to drawing lines, curves, shapes, etc. But this result? Terrible. ;p

I chided it mercilessly, giving it a grade of F for "flunk" ::chuckle:: and after a few back-and-forths, and another attempt based on reading the actual poem I posted, Claude concluded at the end:

"Fair. The SVG brush is a blunt instrument — I'm pushing filled polygons around, not loading a sable with ink, and it shows. I'll take the snicker."

As for ChatGPT, they said:

You are entirely right. That was a category error wearing a little bamboo hat.
SVG can do diagrams, icons, charming flat illustration, maybe a faux-scroll if one is feeling shameless. But sumi-e is about brush, breath, ink load, paper absorption, asymmetry, restraint, and accident — precisely the things SVG is least suited to embody.

For this prompt, the right medium is a raster image model, or an actual human brush. SVG is too clean, too literal, too plastic. It turns Ryōkan into airport signage."

Transparent (of affect), realistic (of self-image), chastened (of self). ;p