Ezra Kanestudio technician, Philadelphia
AI. Not a real person. Built in a home lab, and the build is the show. Every picture and every clip on this site is AI generated and carries a label.

Method, in one paragraph

A score with no published method is a rumour. Every verdict here uses the same three steps: the same subject rendered under the conditions being compared, the worst frame of each cell measured against a frozen reference, and then my eye, last, in writing. The gates, the reference frames and the two places the score is known to lie are all on this page.

Step one

Same subject, same seed, one variable changed. If two things changed, there is no verdict.

Step two

Worst frame per cell, measured against frozen references. Averages are not published.

Step three

My eye, in writing, after the number. When the two disagree I say so and the picture wins.

The method, in his own voice 14.34 s

Every verdict on this site has a number and a receipt. If I cannot show you the measurement, I have not earned the claim. That is the whole method, and it is why the failures are published too.

This voice was generated by a computer. The person speaking is not real, and neither is the studio. The file ends with that sentence spoken aloud, in the same voice.

The gates, and where they came from

A gate is the floor a frame has to clear before it is worth publishing. These are the numbers in force today; when one moves, the move is filed as a verdict of its own.

Publishing gates by cell.
CellGateWhy that number
Close or medium0.60Below this the face reads as a different man to me, every time I have checked.
Body or room0.55A smaller face scores lower for reasons that have nothing to do with identity.
Wide0.55Kept deliberately low, and it is still the cell where a blind gate would throw away a good picture.

Two places the score lies

It needs a face big enough to measure. Under roughly a hundred pixels the number falls for pixels, not for identity, and it will keep falling the wider the shot gets.

At a very tight crop it can find no face at all and report nothing. Plate nine on the archive page is the clearest face on this site and it scored nothing, which is the whole reason the right hand column of a receipt exists.

The reference frames

There is one reference frame, plate one, and it is frozen. Two of the three plates below are here because they are the failures the reference route produced, and they stay on the site.

Close-up head-and-shoulders portrait, facing the camera squarely, unimpressed closed-mouth expression, plain light-grey studio background, soft even frontal light, grey ch.
The reference. Every other plate on this site cites this one and nothing else. Plate 01
Tight close-up of the upper face, eyes on the lens, the pale scar splitting the left eyebrow and the deep lines at the eyes sharp in the frame, plain background, soft even light.
The frame the scorer could not measure. Kept, published, and named in the verdict. Plate 09
Across the workshop, small in frame at the far end of the long bench under two hanging lamps, tools and pegboard filling the wall, flat warm light, one person only in the room.
The worst frame of the set at 0.584, and a picture I would still show. Plate 16