Flipbook renders every browser page as a real-time AI-generated image; navigation works by clicking image regions, with no HTML or code layer underneath.
Key Takeaways
Every page is a pixel-rendered image; text, layout, and navigation hotspots are all produced by the image model, no HTML layer.
Clicking any image region generates a new image exploring that element in more depth; the entire navigation graph is computed on demand.
Content sources from agentic web search plus model world knowledge; expected accuracy is comparable to ChatGPT, Gemini, or Claude.
Live video stream feature combines a custom optimized video generation model with the image system to animate pages and create seamless transitions.
Roadmap: real data integration, interactive actions, and data storage inside Flipbook; goal is replacing separate apps and websites entirely.
Hacker News Comment Review
Reliability split: one commenter reproduced a detailed car-suspension torque spec diagram with accurate figures and clickable drill-down components; another got fabricated and incoherent outputs on basic counting and an emoji’s history.
Gemini 429 rate-limit errors surfaced directly in the UI during the traffic spike, confirming Gemini as the image backend and making per-query inference costs visible to end users.
Skeptics frame it as too unreliable for any factual use case; defenders see the click-to-drill-down image paradigm as closer to a living illustrated reference manual than a chatbot interface.
Notable Comments
@giobox: generated correct torque specs for a specific car suspension with clickable components; called it “a living version of a classic illustrated Haynes workshop manual.”
@totallygeeky: fabricated the seahorse emoji backstory entirely and produced incomprehensible counting diagrams; argues unreliability makes the product only usable if you cannot detect the errors.
