cursed_browser
True AI-Native Browser — a VLM reads the HTML and hallucinates the page.
What is it?
What it is
Cursed Browser is an AI‑native web rendering engine that asks a large language model (LLM) to parse a page’s HTML token‑by‑token and then uses a visual‑LLM (VLM) to hallucinate the pixel output. Each page load produces a unique, artistic rendering rather than a deterministic layout.
Why it exists
It demonstrates how LLMs can replace traditional browser engines by interpreting markup and styles through next‑token prediction, eliminating the need for heavy parsing logic. It offers a lightweight, creative alternative to conventional browsers, showcases the power of AI for web rendering, and removes unnecessary dependencies on the web during development.
Who should use it
AI enthusiastsexperimental tech userscreative developersartists interested in AI-generated visualsresearchers in AI-native applications
Who should avoid it
users requiring stable/accurate rendering for productivitynon-technical audiencesthose avoiding experimental AI tools
How it works
A quick walkthrough in plain English
How cursed_browser works
Step 1 of 3
You interact with it
Open cursed_browser, send a request, or connect it to your stack.
Features
Advantages
- Unique AI-driven rendering approach creates artistic outputs
- No reliance on traditional web infrastructure (model memorizes web content)
- Highly efficient V2 architecture with minimal bloat
- Passes rigorous Acid3 compatibility test
- Creative and unpredictable visual interpretations
Disadvantages
- Potential for hallucinated or inaccurate visual representations
- No formal license specified (may affect adoption)
- LLM dependency could introduce latency or unpredictability
- V2 roadmap suggests high resource requirements for enterprise use
- Humorously framed
Installation
FAQ
What is Cursed Browser?
Cursed Browser is a rendering engine that uses a visual language model (VLM) to look at a page's HTML and draw what it thinks the page looks like, making each render a unique, AI‑generated artwork.
How does Cursed Browser render web pages?
It feeds the HTML to an LLM which interprets the markup token‑by‑token, predicts CSS via next‑token prediction, and then uses a VLM to hallucinate the pixels that represent the page.
How does Cursed Browser compare to other "AI Native" browsers like Arc, Dia, Comet, and Atlas?
Unlike those browsers, Cursed Browser parses HTML with an LLM token‑by‑token, interprets CSS via next‑token prediction, hallucinates pixels with a VLM, and even upgrades all hyphens to em—dashes—features none of the others provide.
What are the current milestones in the project's roadmap?
V1: An LLM looks at HTML and draws what it thinks a browser would show. V1.1 (current): Removes unnecessary web dependency by relying on the model's pretrained knowledge, making HTTP fetch a formality. V2 (planned): The LLM will write a brand‑new browser engine from scratch for each page load, supporting only the features needed.
Where can I see examples of Cursed Browser's output versus a traditional browser?
The README includes side‑by‑side image comparisons for Wikipedia, Hacker News, cbc.ca, Acid 3 test, a Google search for 'cursed browser', and the scosman.net homepage, showing Cursed Browser's render on the left and Safari's render on the right.
Featured in Videos
YouTube tutorials and walkthroughs for cursed_browser
Alternatives
Similar projects ranked by category, topics, and text overlap.
