I tried the studio ghibli prompt on a photo my me and my wife in Japan and it was... not good. It looked more like a hand drawn sketch made with colored pencils, but none of the colors were correct. Everything was a weird shade of yellow/brown.

This has been an oddly difficult benchmark for Gemini's NB models. Googles images models have always been pretty bad at the studio ghibli prompt, but I'm shocked at how poorly it performs at this task still.

Could be they are specifically training against it. There was some controversy about "studio ghibli style". Similarly how in the early days of Stable Diffusion "Greg Rutkowski style" was a very popular prompt to get a specific look. These days modern Stable Diffusion based models like SD 3 or FLUX mostly removed references to specific artists from their datasets.

You might try it again with style transfer: 1 image of style to apply to 1 target image

I wonder ... do you think they might not be chasing that particular metric?