Controlling Perspective and Viewpoint in Prompts

Perspective is where the viewer is standing and how the space converges from there. It is one of the strongest levers in a prompt, because changing it changes everything else — what dominates the frame, what the subject feels like, what you can even see. And unlike most visual qualities, it has proper vocabulary that models understand well.

Height and Angle First

The simplest control is where the camera sits relative to the subject. Eye level is the default and mostly gets assumed if you say nothing. Low angle, looking up, makes things loom and reads as power or threat. High angle, looking down, shrinks the subject and reads as vulnerability or overview.

The extremes have their own names and they work. Worm’s eye view puts you on the ground looking almost straight up, with heavy convergence in the verticals — buildings leaning inward, a figure towering. Bird’s eye view puts you well above looking down at an angle. Top-down, overhead or flat lay means straight down, ninety degrees, no convergence at all, which is a genuinely different geometry rather than a steeper bird’s eye. If you want the flat-lay look, say overhead or top-down; “bird’s eye” will keep giving you a tilt.

Do not stack conflicting terms. “Low angle overhead shot” is not a viewpoint, and asking for it gets you whichever half the model weighted more.

Vanishing Points and Constructed Space

Naming the perspective construction gives you control over how the space converges. One-point perspective means all the receding lines run to a single point in the middle — corridors, tunnels, railway tracks, a road straight ahead. It is the most recognisable and the most reliably produced. Two-point perspective puts you at the corner of something, with lines running off to both sides, which is the natural look for a building seen from a street corner.

Three-point adds a vertical vanishing point and is what you actually get with an extreme low or high angle. Naming it explicitly reinforces that lean rather than fighting it.

Isometric is the opposite request: no vanishing point at all, parallel lines stay parallel, everything the same size regardless of distance. It is well understood by models and produces that clean diorama look familiar from game art and technical illustration. Related terms — orthographic, axonometric, flat projection — pull in the same direction. Note that isometric fights hard with photographic language; ask for an isometric photograph and you will get a muddle.

Forced Perspective and Lens Distortion

Forced perspective is a trick, not a geometry: something near the camera made to look the same size as something far away, the tourist holding up the leaning tower. Ask for it by describing the trick — a hand in the foreground appearing to hold a distant building — rather than trusting the term to carry the idea alone.

Wide-angle and fisheye descriptions exaggerate convergence and bulge the edges, which is a strong stylistic choice more than a viewpoint. They pair well with low angles and badly with anything that needs to read as measured or neutral. If the geometry is the point of the image, keep the lens language minimal and let the viewpoint terms do the work.