For the sake of interest, I trained the hypernetwork on the same dataset as LORA. The result was extremely interesting and beautiful.

I used the same promts for clarity.

In general, the hyper network works well with values of 1, but also with values of 0.5 is also not bad.