A new project called HMML (HyperMedia Markup Language) proposes treating HTML as a native image format instead of generating pixels. The system packages markup, images, and animations into a single binary file that decodes natively in browsers. Rather than flattening everything into one raster, HMML keeps vector, text, and raster elements separately addressable.

The technical approach uses a 2.5 KB reader that achieves roughly 830 MB/s decode speeds. Files come in about 25% smaller than base64 encoding while maintaining zero dependencies. HMML employs a PNG-style signature with self-describing chunks - markup stays as text while images remain raw bytes. Unknown chunks are safely skipped, allowing the format to evolve without breaking compatibility.

For developers, this shifts the workflow dramatically. Instead of hand-wiring layout, styles, and assets around AI-generated images, the model ships complete scenes as one portable file. Applications fetch the .hmml, decode it, and mount directly into a div. The approach treats generated output as source code rather than frozen screenshots.

Whether models adopt this directly remains uncertain, but the core idea addresses a real tension: current image formats crush diverse media types into identical rasterized blobs. HMML un-conflates these by preserving their native formats within a single distributable package.