A frame from a typical GRMHD model (MAD, a_* = 0.5, i = 30^∘, R_high = 160) convolved with a 2D Morlet wavelet, showing which image-domain features contribute to the visibility amplitude at ρ = 4 Gλ (left), 10Gλ (middle), and 30Gλ (right), filtering for horizontal (top) and vertical features (bottom). The original frame is shown in gray behind the convolution. As expected, large-scale features dominate the low-frequency (u, v) data, while small-scale features dominate the high-frequency (u, v) data.