25Prompts beyond code

Code isn't the only reason to have an AI, and the rules differ.

Text. The main problem is the smooth averaged voice you can spot instantly. Examples cure it: give it a piece of writing you like and ask for the same. And ban clichés explicitly - "no throat-clearing openers", "don't start with 'in today's world'". It works.

Images. The opposite: the more detail the better. What's in it, the angle, the light, the style, the background. A short prompt gives an average result. And it nearly always helps to say what should not be there. By the way, the newest models can finally put readable text on images; that used to be a constant pain.

Video. Still the most temperamental area, and it runs on "lots of attempts". Describe the scene like a director: what's in frame, what happens, how the camera moves. Good news: in Gemini Omni you can fix a failed clip right in the conversation instead of generating it again from scratch.

Common to all of it: small test first, big run after. Check the style on one paragraph before asking for ten pages.