Drag the picture to pan and tilt the camera. Use the sliders for lens, height and subject position; the score rewards classic composition rules.
Pinhole camera. A point at camera coordinates (X, Y, Z) lands on the sensor at x = f X/Z, y = f Y/Z. The image is a 36 × 24 mm frame, so the frame position is u = 0.5 + fX/(36 Z), v = 0.5 − fY/(24 Z). A subject of height H at distance Z fills the fraction f H/(24 Z) of the frame height. Pan and tilt rotate the camera before projecting.
Horizon. The horizon is the image of directions at infinity: vh = 0.5 + (f/24) tanθ for tilt θ (up is positive), so a long lens moves it faster. Parallel lines (the road edges, fence rails) meet at a vanishing point on the horizon — they lead the eye there.
Dolly zoom. Doubling both f and Z keeps the subject the same size (f/Z unchanged) but doubles the size of far objects such as the mountains: a longer lens compresses perspective.
Grids. Rule of thirds lines at 1/3 and 2/3; golden grid at 1 − 1/φ = 0.382 and 0.618 with φ = (1+√5)/2. The four intersections are "power points".
Score (heuristic). Subject: exp(−(d/0.07)²) with d the distance to the nearest power point. Horizon: same form (width 0.05) for the nearest third or golden line. Balance: visual mass m = c √A (area fraction A, contrast weight c), the see-saw offset |∑m u/∑m − 0.5| should be small. Leading lines: distance from the vanishing point to the subject. Weights 30/25/25/20 %.
Honest simplifications. Aesthetics are not physics: the contrast weights and widths are my choices, only the geometry is exact. No lens distortion, no depth of field, no roll, flat-coloured scene.