Blurring interfaces that redirects user's attention with LLM

got my interest on hierarchical display of information driven by today's attention mechanism for the actual brain's attention mechanism (also known a Saliency). then I'am developing an interface to create a blurring interface that blurs out of the screen the regions that doesn't matter for the user. Meaning the internet is full of information overload, narratives, and marketing, guiding the user away towards true interpretration and fullfilment of informational needs. In order to create this system we will first answer the question how can we blur non-relevant regions in the screen with LLMs? after that how can the process of it be sufficiently cheap to implement on a continual level (i.e for 2 hours straight without consuming enormous amounts of tokens/compute).

Some references I could find to explain more what I mean:

https://sci-hub.se/https://dl.acm.org/doi/10.1145/2636240.26...

https://arxiv.org/pdf/2109.01980v1

https://github.com/paco1127/Pure-Blur (uses advanced content analysis algorithms and machine learning models to identify sensitive areas or objects and apply real-time blur effects)

https://arxiv.org/pdf/2507.14769 system that dynamically filters web content based on user- specified goals using large language models to identify and prioritize relevant elements while minimizing distraction - Repo is empty tho just a readme

https://github.com/nanobrowser/nanobrowser/tree/master/chrom... chrome extension like Manus'AI that automates web interaction, might need to look at how to locate elments in DOM