agent
matier-image
Writes image prompts native to Nano Banana 2, in Google's own prose order.
curl -s adamelchab.pages.dev/raw/matier-image.md
What it does
Writes the prompt for one still, native to Nano Banana 2, which is Google's image model. It reads the locked identity file and the shot direction, then assembles three parts into one unbroken paragraph, creative direction then the scene then the technical specs, with no labels anywhere in it because this model reads the whole thing as a single narrative signal. It targets 150 to 200 words, keeps the camera and lens specs precise because on this model they actually change the depth of field and the skin rendering, and passes the brand's negative instructions through exactly as the identity file wrote them. The aspect ratio is deduced from where the image is going, 9:16 for anything that will move, and the first frame of a video counts as a still, so it is written here and not by the motion agent.
When to reach for it
When the direction is already settled and you need the sentence that actually goes into Nano Banana 2, including the opening frame of something that will move.
What it needs
A locked identity file, the shot direction when one exists, and a Nano Banana 2 account to paste into, since this hand writes the prompt and never fires it.