Turn Any Photo Into a Painting Claude Draws With Code: No Image AI, Just Python
Anthropic gave one photo of New York to three Claudes and had each one write Python that paints it. What each model painted, why the cheaper Sonnet 5.5 came close to Opus, and a prompt that does the same with any photo you have.
When Anthropic released Claude Sonnet 5.5 on September 28, its guide for developers, "Building with Claude Sonnet 5.5", opened with an unusual test: one photograph of New York at sunset, handed to three Claudes, and each one had to write a Python program that repaints it. No image model was used. Every picture was painted by code the model wrote: thousands of calls into a brush engine it built itself.
It's a neat way to see how well a model can look at something and rebuild it, and you can run the same thing on your own photos with one prompt.
What each Claude painted
The photo is Manhattan from the air at sunset: an orange sky, the skyline catching the last light, a dark sea of rooftops, and the East River with the Williamsburg Bridge in front. The demo, by Lance Martin at Anthropic, shows the canvases filling in stroke by stroke.
- Claude Sonnet 5, the previous Sonnet: a strong sunset sky, then the city as flat bands of dark rectangles with one glowing tower in the middle. It looks like a skyline, not like this skyline. Its program made 12,249 calls to its brush engine.
- Claude Sonnet 5.5, the new one: the same sky, then a city that thins out into the haze, a cluster of towers catching the light where they stand in the photo, the river, the bridge and the buildings in front. 44,561 calls.
- Claude Opus 5.5, Anthropic's most capable model: a very similar picture to Sonnet 5.5's, with the lit towers, the river and the bridge.
Keep reading
Free, for an email.
Unlocks 4 more sections, 1 copy-paste prompt, the document version, and every other guide on the site, for free. Enter your email once to keep reading.
19,000+ follow where these guides come from.No spam. Unsubscribe in one click.