I imagine that's because LLMs are of most interest to the Hacker News crowd: they can help write code, and you can build systems on top of them that can "understand" and respond in human language.
Generative image / video / audio models can produce output in image, video and audio. Those have far less applications than models that can output text, structured data and code.
HN is by ICs who write code, the 90% of the folks that build all the stuff, largely neutral to negative. It has gained some excellent traction with 10% of the folks, but it is quite behind compared to ai coding subreddits. Months behind.
Generative image / video / audio models can produce output in image, video and audio. Those have far less applications than models that can output text, structured data and code.