Composition Prompts: How I Direct the Frame Without Micromanaging It
Composition cues in prompts steer framing without strangling the model. Here is the vocabulary I use and the ones I avoid. Composition is where prompts either help the model or strangle it. Too little composition direction and the model defaults to centered, symmetrical framing that reads as generic. Too much composition direction and the model stops rendering freely and produces stiff, awkward arrangements that look forced. The middle ground is a small set of cues that tell the model where to place weight in the frame, then let it solve the details. This guide covers the composition vocabulary I actually use and the pitfalls I avoid. Working From the Subject Outward I write composition cues after the subject and action are settled, never before. Starting with composition makes the prompt read as a layout description rather than an image description, and the model sometimes renders a layout instead of a scene. "A cyclist on a country road, leading line of the road pulling the eye to the right side of the frame" works. "Leading line to the right side of the frame with a cyclist on it" produces something more diagrammatic. Subject first, composition second, always. The composition cues I use describe placement, movement, and depth. Placement tells the model where the subject sits in the frame. Movement tells the model where the eye should travel after the subject. Depth tells the model how the foreground, midground, and background relate.