A structured prompt builder for the new generation of models — image, video, world-action and chat. Compose subject, camera, lighting, style and constraints deliberately instead of hoping a string of adjectives works. Everything is local; history stays in your browser.
Target
Scene
Task specification
Prompt structure
Prompt
—
Chips toggle on/off; multiple selections are ordered as clicked.
History (local)
Last 12 prompts, stored only in this browser.
Why structured beats a wall of adjectives: generation models respond to compositional structure — subject, camera, light, style are different control axes, and mixing them into one clause makes them fight. This tool keeps the axes separate and assembles them in the order models weight most. The action-task mode exists because the newest "action models" are prompted with goal / scene / constraints / success criteria, not prose. Nothing here is magic; it is a discipline bench. No model calls are made — you take the output to whichever service you use.