Prompt Fidelity and Output Drift

The VTL Kernel is a deterministic measurement device for the spatial organization of visual mass in a 2D image. It does not evaluate aesthetic quality, subject matter, or style. It measures where gradient energy lives in the frame, using Sobel operators to extract edge structure and then computing nine independent metrics from that gradient field.

What it measures: Every pixel with significant contrast, texture transition, or edge contributes to a mass map. The kernel then describes that mass: where it sits, how spread out it is, whether it forms a single body or many fragments, how strongly it pulls toward the frame edges, and what directional axis runs through the composition.

What to take away: The nine metrics are coordinates, not grades. A centered mass is not inherently better than an offset one. That depends entirely on intent. The value of the kernel is in making spatial habits visible and measurable, so you can decide whether the composition is doing what you want it to do, or settling into a learned default.