Claude skips the training set: HTML/CSS rendered in a headless browser makes his site images

shashib · x · 2026-09-09

The author argues "AI-generated images" are two different things: diffusion models reconstruct from patterns learned across billions of scraped photos, versus code-written pages rendered by a browser with no photo dataset involved. His site's hero images and LinkedIn cards use the second kind: Claude writes Python and HTML/CSS against a fixed design system, a Playwright headless browser renders it, and PIL downsamples the capture. No training images anywhere in the path; open-ended illustration is a different matter.

Original post →

More from coding & agent

coding & agent channel →