From 300 Generations to a 12-Image Campaign
If you have 300 AI-generated images and need a final campaign, don't try to pick the final twelve immediately. reduce the collection in stages, first removing obvious failures, then repetitive generations, then judging the remaining images as a campaign rather than 300 individual files.
We're going:
300 → 120 → 50 → 25 → 15 → 12.
Not because twelve is magic.
We just have to stop somewhere.
300 → 120: The First Cut
Don't find the best images.
Find the obvious no's.
Bad anatomy, wrong product, weak composition, images outside the visual world, accidental weirdness that isn't the good kind of weirdness.
Gone.
Fast.
120 → 50: The Cousins
Now similar images have to fight.
Seven versions of essentially the same portrait?
Pick one or two.
Five images doing exactly the same job?
They can't all stay because they're pretty.
This part is annoying.
Do it anyway.
50 → 25: The Clusters
Now look for the actual campaign.
Which images naturally belong together?
Which visual direction feels strongest?
What's becoming repetitive?
You're not evaluating isolated frames anymore.
You're trying to see the shape of the work.
Stop
Before cutting again, look for the gap.
Do you have a hero?
Do you have environment?
Close-up?
Product?
Detail?
Movement?
Something unexpected?
This isn't a mandatory shot list. Maybe your campaign intentionally has zero product close-ups.
Fine.
But that should be a decision, not something you discover after delivery.
If there's a real gap, generate for the gap.
Then come back.
25 → 15: The Edit
Sequence them.
What opens?
Where do you get close?
Where does the campaign breathe?
Which images suddenly feel weaker when they're sitting next to each other?
This is where you're editing a campaign instead of ranking pictures.
15 → 12: Make the Cut
Three have to go.
Maybe they're gorgeous.
Sorry.
The final campaign isn't a trophy case for every image you liked.
Look at the twelve.
Does it feel finished?
Good.
Please stop prompting.