Why the same person gets different AI scores

The short answer

The AI scores the photo it is given. Change the light, the angle, the distance or the crop and you have given it a different input, so you get a different output. A score moving between two photos of the same subject is expected, and it says something about the photos.

The useful move is to work out which change did it, and the six-axis breakdown is how you do that.

Photos of one subject vary more than you think

This is a well-studied effect with faces. In Jenkins, White, Van Montfort and Burton (2011, Cognition), the authors report that within-person variability across photos is often very large compared to the differences between people. Viewers meeting unfamiliar faces treated photos of the same person as different people.

The study is about faces, not this scan, but the lesson transfers: one photo is a sample, not a portrait. Expect a range, as what your score means puts it, rather than a single true value.

What actually shifts

Light is the biggest. Side light shows texture, and a hard front flash erases it, which hits the skin texture axis directly. Angle is next: shooting down foreshortens the subject and changes what the shape and proportions axis reads.

Crop and distance come after. Close shots stretch near parts and shrink far ones, and a cluttered frame can cost you on the two subjective axes, overall appeal and erotic impact. File quality closes the list: a screenshot or a heavily compressed copy loses detail the scan would otherwise read.

How to find out which change did it

Change one thing at a time. Keep the subject, the setting and the distance, vary only the light, scan, then vary only the angle, scan again. Because the breakdown scores each axis separately, you can see which axis moved with which change instead of guessing from the total.

The photo guide covers each of these changes in how to take a photo that rates accurately, and how AI cock rating works explains what the axes are reading.

What to hold constant

To compare fairly, change one variable and freeze the rest. Prop the phone or lean it on something so the distance and angle stay put, shoot in the same spot at the same time of day so the window light matches, and keep the same crop. Then vary only what you are testing.

Most people change three things at once without noticing, for example a new room, a new time and a new angle, and then cannot say what moved the score. Holding things constant takes a few minutes and turns a guess into a reading.

A worked comparison

Imagine two photos of the same subject: one in daylight from a window, level, from a couple of steps back, cropped in; one under a ceiling light, from above, close up. You would expect the second to come back lower on skin texture and on shape and proportions, because those two axes read exactly the things that photo got wrong.

That is a hypothetical, not a promise, but it is the pattern to look for. If the lower axes line up with the changes you made, the photo explains the difference and nothing else needs explaining.

Reading a drop and a rise

A lower second score does not mean you got worse, because nothing about you changed between the shots. Equally, a higher one does not mean you improved. Both describe the photo. The useful reading is which axis moved, and whether the change you made is the likely reason.

If you are comparing photos to decide what to post, compare the breakdowns, not the totals. The six rating axes goes through what each one reads, so you can tell which change did the work.

Which score to keep

Keep the one from the cleanest photo, not the highest number. If you post to the leaderboard, the entry is tied to that rating and its breakdown, so a clean photo gives you a result you can stand behind. How to climb the leaderboard covers what to do with it from there.

When to stop re-scanning

Once two clean photos land close together you have found your range. Past that point, further scans mostly sample noise. The scan is free with an account, but a higher number from the nth attempt tells you about the attempt, not about you. Keep the best clean one, and if you want to go further with it, AI judges will render a clip from the photo you chose.

It also helps to write down what you changed between shots. A one-line note per photo, such as window light or camera lowered, is enough, and it saves you from remembering after the fact which of three similar photos produced the number you liked. Memory is a poor instrument for this, and the breakdown only helps if you know what to compare it against.

None of this needs special equipment. A phone leaned against a book and a window are enough to run a fair test, and the whole experiment takes a few minutes.