Part 01
Sixty-nine images, ten slots
The Atompunk Hotrods series left me with 69 finished images, and an Instagram carousel only takes ten. So I asked my AI assistant to score all of them with a table: nine criteria, each from 0 to 5, such as technical quality, composition, physical logic and impact. With the total it sorted them into a tier list: S for the best, then A, B, C and D, and X for the ones that are not valid. Then I moved the images between folders myself.
Krea 2 · Atompunk Hotrods · score 95.96 out of 1005 attempts
The table's top five
These are the five highest scores. The table put fourteen images in tier S. I kept one of those fourteen. I moved eight down to A, because next to the finalists they had small details. The other five I moved further down.
- #1Option
The table's highest score. I moved it down three tiers, from S to C.
Score 95.96 · tier S → mine: C - #1My pick
Tied for first place. The only one of these five that made my ten.
Score 95.96 · tier S → mine: S - #3Superseded
Very good. It was left out for small details when I put it next to the finalists.
Score 94.54 · tier S → mine: A - #3Superseded
Another one I gave up for small details. It is still one of the strong ones.
Score 94.54 · tier S → mine: A - #5Option
The table put it at the top. I moved it down two tiers, to B.
Score 93.48 · tier S → mine: B
10 attempts
My ten
These are the ten I chose for the carousel. Eight came from tier A and one from B. Looking at them together, I see what the table was not counting: in almost all of them she stands next to the car and you can see the whole car, and no two share the same pose, color or angle.
- #43My pick
For the table it was number 43 out of 69, in tier A. For me, one of the ten.
Score 85.3 · tier A → mine: S - #26My pick
For the table it was number 26 out of 69, in tier A. For me, one of the ten.
Score 88.72 · tier A → mine: S - #34My pick
For the table it was number 34 out of 69, in tier A. For me, one of the ten.
Score 87.82 · tier A → mine: S - #43My pick
For the table it was number 43 out of 69, in tier A. For me, one of the ten.
Score 85.3 · tier A → mine: S - #34My pick
For the table it was number 34 out of 69, in tier A. For me, one of the ten.
Score 87.82 · tier A → mine: S - #39My pick
For the table it was number 39 out of 69, in tier A. For me, one of the ten.
Score 86.12 · tier A → mine: S - #62My pick
The table had it at number 62 out of 69, in tier B. I moved it up two tiers.
Score 76.44 · tier B → mine: S - #15My pick
For the table it was number 15 out of 69, in tier A. For me, one of the ten.
Score 89.94 · tier A → mine: S - #26My pick
For the table it was number 26 out of 69, in tier A. For me, one of the ten.
Score 88.72 · tier A → mine: S - #1My pick
Here we agreed: the table had it at the top, and so did I.
Score 95.96 · tier S → mine: S
What I learned
The table is good for sorting and for catching what is broken. It is not good at choosing. Of its fourteen favorites, only one made my ten.
With a fixed number of slots, “very good” is not enough. I moved eight strong images down just for small details.
For a carousel, variety weighs more than the score of each image. I want different poses, colors and angles next to each other.
Two images looking alike is not a reason to mark one down. The table penalized four for repeating an idea, and I moved all four up.
My change wins, and I keep it written down with its reason when I have one. That way the next proposal looks more like my taste.
Follow the lab
Every experiment goes up as it happens: the picked images, the clips in motion and the batches that didn't work.
Write to me on Instagram