Colour temperature is an input variable

A camera does not record colour the way a light meter would. It guesses what kind of light it is looking at, then shifts every pixel to correct for that guess, and the guess is not always right.

What white balance is actually doing

Auto white balance estimates the colour temperature of the ambient light - warm and orange like a tungsten bulb, cool and blue like overcast daylight, greenish like some fluorescent tubes - and then applies a correction so that things the camera expects to be neutral, like a wall or a shirt, come out looking neutral. This is an inference, not a measurement of the actual light source, and it runs on every photo whether or not anyone notices it happening. When the estimate is close, the correction is invisible. When it is off - mixed lighting, an unusual bulb, a coloured wall reflecting light back onto the subject - the whole frame gets recoloured in a direction that has nothing to do with the subject and everything to do with the camera's guess.

Why that reaches the score

Skin tone and skin texture are exactly the properties a scoring model is built to read, and both are colour-dependent in ways that white balance can shift on its own. A slightly warm-shifted photo can make skin look more flushed or more even; a cool-shifted one can make it look paler or bring out texture the warm version smoothed over. None of this is the skin changing. It is the recolouring changing what the model receives as skin. Lighting direction has a bigger, more structural effect on scores because it changes shading and contour rather than colour, but white balance is the variable that moves the same photo's colour statistics without touching the light's direction or intensity at all - a distinct and often overlooked source of the same kind of noise.

What to take from this

Two photos taken back to back, in the same spot, under the same light, can still come out with visibly different colour casts if the camera's white balance estimate is inconsistent between shots, which some phones are prone to under mixed or unusual lighting. That is one more entry on the list of things that can move a score between attempts that look, to the eye, like the same conditions - the fuller list is covered separately, and white balance belongs on it as a colour-only variant of the lighting problem rather than a new category of its own. Reading a single result with that kind of noise in mind, rather than treating each score as final, is the habit Penis Rater writes about from the user's side.

This post is about the mechanism, not about camera settings to change; a specific tool's photo instructions are the place for that, and Rate Cock's photo guide is where they live. Skin tone as a source of systematic bias across a training set, rather than as a per-photo variance source, is a different and more serious question - that side of it is covered elsewhere and is not what this note is about. A human reviewer sees the photo with their own colour perception rather than through a camera's correction pipeline at all, which is one of the plainer differences Rate Penis's judging notes touch on, and anyone trying to hold conditions steady across repeated shots for comparison purposes will find Measure My Cock's gear notes cover consistent lighting setups from the equipment side.