On-camera flash removes the shadows the model uses

Direct flash flattens a photo because it removes the shading a model uses to infer shape. A model never sees three-dimensional form directly; it reads the gradual darkening across a curved surface as it turns away from the light.

On-camera flash removes most of that gradient in a single frame, so a model reading a flash photo is working with less shape information than it would from almost any other light source.

What flash does to a shaded surface

A flash mounted at or near the lens sends light straight along the camera's line of sight, which means every surface facing the camera is lit almost equally regardless of how it curves. Light coming from a direction other than the camera's own position produces a genuine gradient across a curved surface, and that gradient is a large part of how a model - or a person - reads shape from a flat image in the first place. Computer vision has formalised this since Berthold Horn's 1970 MIT report on shape from shading, which recovers the shape of a smooth object from one image by relating brightness to surface orientation - given knowledge of where the light source is. Flash collapses that gradient toward flat: the surface is bright everywhere it faces the lens, dark everywhere it does not, with almost no gradual transition in between. The result looks harsher to the eye, and it is also a genuinely different, less form-revealing input to whatever is reading it downstream.

Why this reaches shape-related scores specifically

Axes that respond to shape, proportion and contour are, structurally, reading shading gradients as their main source of evidence, because that is the only channel a single 2D image offers for three-dimensional form. Take away the gradient and the model is inferring shape from a flatter, more ambiguous signal than it had with softer, angled light. This does not make the model refuse to score, and it does not necessarily push the score in one consistent direction - what it does is remove a useful cue, and a model working with less useful signal tends to produce a noisier, less predictable result on exactly the axes that depend on that signal. Presentation-adjacent axes, less dependent on inferred 3D form, are comparatively less affected, because they were reading surface and framing quality rather than contour to begin with.

What this is not

This is not shooting advice, and there is no recommendation buried in it - the mechanism is the point, not what to do about it. A specific tool's photo instructions belong to the tool: Rate Cock's photo guide is where that lives. Flash is one entry in the wider set of things that shift a score between attempts that otherwise look similar - the fuller catalogue is covered separately - and it sits specifically alongside lighting direction as a shading-cue problem rather than a colour or framing one. A human reviewer looking at a flash photo can usually tell the shading is flattened and discount for it consciously, something a scoring model has no equivalent step for - Rate Penis's judging process describes a read that works that way. None of this bears on a physical figure either; Measure My Cock's method does not depend on shading at all, since it starts from a tape rather than an inferred contour. Penis Rater covers, from the user's side, how much weight to put on a single shape-axis result once you know what produced it.