I gave Opus 5.5 one prompt and six hours to visualize Invisible Cities
Points and comments are a snapshot, not live.
Claude Opus 5.5 generates a mesmerizing interactive 3D visualization of all 55 Invisible Cities from a single prompt.
The author gave GPT-6 Astra and Claude Opus 5.5 the same prompt: build a three.js visualization of all cities from Italo Calvino's book. GPT-6 Astra produced a decent result in 53 minutes for about $10. Claude Opus 5.5, despite ignoring the request to use 6 hours and finishing in 1 hour 25 minutes (with 6 parallel subagents, ~$74), generated an interactive city viewer the author calls mesmerizing. The author notes the model's design capability feels like a leap beyond previous models, but also wonders about their own role in creating interactive media as AI capabilities advance.
The Lens Lab interactive optical demo by Ryan Sael is cited as the inspiration: a 3D camera lens simulation built with Opus 5.5 in one shot for $25.66. The author previously struggled with earlier models for interactive visualizations, requiring hand-holding and fixes for AI slop and overlaps.
What commenters are saying
Many commenters urge readers to experience the book first before viewing the AI-generated cities, arguing the visualizations miss Calvino's point and corrupt the reader's own imagination. Some find the AI cities beautiful at a glance but hollow on inspection, pointing to nonsensical bridges and details that break under scrutiny.
One commenter contrasts human-made art, where zooming in reveals intent and hidden details, with AI output described as a 'cardboard facade with no passion.' Another notes the cost comparison ($10 vs $74) and questions what the actual hardware and training cost should be. A few express wonder at the capability and want the book after seeing the demo.