Flipbook is an experimental “infinite visual browser” released on April 23, 2026. It was built by a small team including Zain Shah, formerly of OpenAI. A conventional web browser receives HTML and hands it to a layout engine for rendering, but Flipbook has no such step. Every image and every character on screen is drawn as pixels, on the spot, by a video generation model. Tap anything that catches your eye on a page and the next page ─ one that digs into that element ─ is generated immediately. Rather than following links, you explore as if walking across a terrain of knowledge. The project has been described as drawing inspiration from Apple’s classic HyperCard.
Key Features
- No HTML, no layout engine: There is no DOM structure like an ordinary web page. The entire page is output from a generative model, making it closer to a continuously painted canvas than to a browser
- Every tap gives birth to the next page: Because no link destinations are prepared in advance, the concept of a broken link does not exist. Point at an element on screen and a new page on that subject is created there and then
- Real-time streaming at 1080p / 24fps: It combines a video generation model from Lightricks (LTX Studio) with serverless GPUs from Modal Labs, delivering the result as video over WebSocket
- Text and illustration are drawn together: Text and figures are not separate elements; they are generated as a single picture, like a page from an illustrated book. Instead of text streaming in as in a chat UI, you receive the whole layout at once
- A small experimental team: Centered on Zain Shah (Y Combinator S13, previously at OpenAI and Samsung), with engineers formerly of Apple, Slack, and Humane
Pricing
| Plan | Price | Notes |
|---|---|---|
| Experimental release | Not disclosed | No paid plans or billing structure have been published. It is offered as an experimental project |
Pricing is as of August 2026. Please check the official site for the latest information.
Pros & Cons
✅ Pros
- You receive information as an illustrated page rather than as a wall of search results or chat text
- You are not bound by whether a link exists ─ any point on screen can become the starting place to dig deeper
- Because the layout itself is generated each time, discovery works differently from browsing existing web pages
- You can put your hands on the idea of using a video generation model as the interface itself
⚠️ Cons
- Since the screen is entirely video, you cannot select or copy text or search within the page. Screen reader support is unlikely, so accessibility falls far short of the ordinary web
- The text shown is model output and may contain typos or factual errors. It cannot be used to verify facts
- Streaming video continuously means heavy bandwidth use and a heavy GPU load on the server side
- It is an experimental project, so continuity and availability are not guaranteed, and no pricing structure has been published
Comparison with Similar Services
| Criteria | Flipbook | websim.ai | Perplexity | Ordinary web browser |
|---|---|---|---|---|
| How the screen is produced | AI generates the screen pixels | AI generates HTML | Searches and summarizes the existing web | Renders existing HTML |
| Output format | Video (a sequence of images) | Web page (HTML) | Text with source links | Web page |
| Links | None (generated on tap) | Generated links | Links to sources | Real URLs |
| Text selection and copy | Not possible | Possible | Possible | Possible |
| Positioning | Experimental prototype | Experimental service | Commercial service | Standard way of browsing |
Who Is It For
- People interested in the new relationship between generative AI and user interfaces
- Engineers and designers who want to try an application of video generation models firsthand rather than watching a demo video
- People who find endlessly scrolling chat-style AI interaction unsatisfying
- Anyone who wants to try research as “looking around and digging” rather than “reading”
Summary
Flipbook is an extreme experiment that replaces the ordinary act of rendering a web page entirely with a generative model. The constraints are significant ─ text cannot be selected, accuracy is not guaranteed, bandwidth use is high ─ and it is not a replacement for everyday research. Even so, it is a rare chance to feel out how far the idea of generating the interface itself can go. Try it briefly and see how the sensation of “reading” a generated screen changes for you.