Real-time Pixel Streaming Demo

Web browsers normally turn HyperText Markup Language, Cascading Style Sheets, and JavaScript into what you see on screen. Zain Shah said his Flipbook prototype skips that step and streams model-made pixels instead. (developer.mozilla.org) (threadreaderapp.com) Shah posted the demo on X in a six-post thread that was live on April 23, 2026, and credited Eddie Jiao and Drew O’Carr as collaborators. He wrote that the team built the prototype to test “every pixel on your screen, streamed live directly from a model.” (threadreaderapp.com) In the same thread, Shah said Flipbook uses a heavily optimized LTX Studio video model, sends live 1080p video at 24 frames per second, and connects over WebSockets to Modal’s serverless graphics-processing-unit infrastructure. WebSockets are a browser feature for keeping a two-way connection open between a page and a server. (threadreaderapp.com) (developer.mozilla.org) (modal.com) That setup differs from standard pixel streaming systems, which usually run an application on a remote machine and deliver its rendered frames through a browser frontend. Epic Games’ Unreal Engine documentation, for example, describes pixel streaming with a packaged app, web services, and an HyperText Markup Language player page. (dev.epicgames.com 1) (dev.epicgames.com 2) Shah said Flipbook’s images can “reshape themselves” to fit a window because there is no fixed layout engine deciding where buttons, boxes, and text belong. He also said any region of the image can become interactive, rather than only elements a developer marked up in advance. (threadreaderapp.com) LTX markets its models as multimodal video generation systems for text, image, and video inputs, and its public materials describe high-resolution, high-frame-rate generation. Shah’s thread suggests his team adapted that kind of model for a live interface stream rather than a pre-rendered clip. (ltx.io) (docs.ltx.video) (threadreaderapp.com) Shah also put limits on the demo. He wrote that Flipbook is “early and slow,” said some clips in the thread were sped up or edited, and said the current version was designed around visual explanations because the models are still limited. (threadreaderapp.com) His public site describes him as a creative technologist at Samsung Design Innovation Center who works on interface design and artificial intelligence projects. That background helps explain why the prototype was framed as a new kind of user interface, not just another video model demo. (zainshah.me) The result is a test of whether a browser can become more like a live video receiver than a document renderer. For now, Shah’s own description keeps it in prototype territory: a working demo that streams pixels first and figures out the rest later. (threadreaderapp.com)