modify a prompt and some tips...

I use gemini flash 2.0 to analyze images (construction plans) the prompts I use give random but promising results I must refine my prompts it's clear and with the number of input tokens available we should be able to achieve it