GitHub - scosman/cursed_browser: A web browser with no rendering engine — the VLM reads the HTML and hallucinates the page.

1 min read Original article ↗

Cursed Browser: Rendering Engine using Visual-LLMs

Cursed Browser asks an LLM to look at the page's HTML and draw what it thinks it looks like. Every page load is a surprise. Every render is a work of art. It's better than correct, it's AI Native.

Examples: Cursed vs Safari

wikipedia rendered by cursed wikipedia rendered by safari

hacker news rendered by cursed hacker news rendered by safari

cbc.ca rendered by cursed cbc.ca rendered by safari

acid 3 test rendered by cursed acid 3 test rendered by safari

google search for 'cursed browser' rendered by cursed google search for 'cursed browser' rendered by cursed

scosman.net homepage rendered by cursed scosman.net homepage rendered by safari

Compared to other "AI Native" browsers

Feature Arc Dia Comet Atlas Cursed
HTML parsed by an LLM token-by-token
CSS interpreted via next-token prediction
Pixels hallucinated by a VLM

Roadmap

V1: An LLM looks at HTML and draws what it thinks a browser would show. Technically a browser. Legally, probably also a browser. Morally, questionable.

V1.1 (current, open-source): Break unnecessary dependency on the web; the model memorized it during pretraining. The live HTTP fetch is a formality, a polite nod to legacy infrastructure. Acid test 100/100.

V2: (oversubscribed $200m pre-seed, enterprise) The LLM writes a brand new browser engine from scratch every time you load a page. No bloat -- the engine only supports exactly the features that page needs. Extremely efficient.