Presets
Playback
Shape
Picture
Camera
Layers
Quality
Model check
The science
The model
Each shape i gets a visual weight w_i = A_i ln(CR_i): its area A_i (circle pi s^2/4, square s^2, equilateral triangle sqrt(3) s^2/4, with s as a fraction of the picture width) times the natural logarithm of the WCAG contrast ratio against the background, CR = (Y_light + 0.05) / (Y_dark + 0.05), with Y the relative luminance from 0 (black) to 1 (white). A shape with the same luminance as the background has CR = 1 and weight 0. The visual centroid is the weighted mean position c = sum(w_i r_i) / sum(w_i), exactly like a centre of mass. The torque about the vertical centre line is M_x = sum w_i (x_i - 1/2); the picture is balanced left to right when M_x = 0. A counterweight of weight w = -M_x / (x_c - 1/2) at distance x_c cancels it.
The seesaw tilt is an analogy, not a measured quantity: the beam of half-length 1/2 turns by atan(M_x / (W / 2)), so a centroid at the picture edge gives 45 degrees. The distance from the centroid to the nearest rule-of-thirds point (1/3, 2/3) or golden-ratio point (0.382, 0.618) is reported as plain geometry, not as a quality score.
Honest limits
Real visual weight also depends on hue, isolation, edges, texture, subject matter and where people look first; this model keeps only size and luminance contrast, which are measurable. It does not judge whether a composition is good art: balance is one property among many, and many strong pictures are deliberately unbalanced. The logarithm of the contrast ratio is a modelling choice (perceived brightness differences grow roughly with the logarithm of the stimulus), not a law.
Independent checks on this page
(1) Torques about the centroid cancel and the parallel-axis theorem holds. (2) Moving all shapes moves the centroid by the same vector; scaling all areas leaves it unchanged. (3) The counterweight solver brings the torque to zero. (4) Contrast and area formulas: black on white is exactly 21, equal luminance gives weight 0, the ratio is symmetric, the triangle area is sqrt(3)/4.
What you see in 3D
Each shape stands on the beam at its horizontal position: footprint = size, height = ln(contrast ratio), colour = luminance. Beam depth shows the vertical position (top of the picture is toward the wall). The yellow ball marks the visual centroid. The beam turns by the analogy angle described above.