Most portfolios that reach us now are technically fine. The lighting is plausible, the composites hold, the grade is competent. That used to be the filter. It is not any more.
Tool proficiency stopped being a differentiator when the tools stopped being difficult. What did not change is that somebody still has to decide what the frame is for. Creative hiring, for us, is almost entirely about that decision.
So we read a portfolio for evidence of judgement, and judgement is mostly invisible in finished work.
We ask about the frames you killed
A reel is a record of outcomes. We want the record of choices. So the first thing we ask for is a set that was rejected — three or four frames good enough to keep, cut anyway — and the reason each one went.
The answers separate people quickly. "It didn't feel right" is not an answer. "The cap highlight was competing with the label, and fixing it would have cost the shape of the bottle" is. Taste is not a mood. It is a set of reasons a person can state out loud, and someone who cannot state them cannot be directed and cannot direct anyone else.
We also ask what a candidate does with a brief they disagree with. The wrong answer is to execute it silently. The other wrong answer is to argue until it becomes their brief. The work belongs to the house that commissioned it, and the job is to make that house's decision better, not to replace it.
Direction is a spoken skill
Our method is called Directed Intelligence, and the claim underneath it is short: we don't generate, we direct. Creative direction is a hiring criterion here before it is a marketing line. A director who cannot articulate a note is a person with opinions.
"Warmer" is not a note. "Bring the key round twenty degrees so the pleats catch on the outer edge" is a note, because it names the thing to change and the reason to change it. We put a frame in front of a candidate and ask for three notes. We are listening for specificity, and for whether the three notes are about one problem or three different ones.
Where AI skills sit
AI skills are assumed and not celebrated. Fluency with a particular model is worth having and worth roughly nothing in two years, because the model will be replaced and the interface will move. We do not test for it.
What transfers is older than any of the tools: light, material, lens behaviour, colour. The useful skill is being able to say why a generated frame is wrong in the language of a camera — the falloff is too fast for a source that size, that reflection belongs to a room that is not in the shot. Someone who can only correct a frame by generating another one is stuck the moment the engine changes. Someone who knows what they were looking at is not.
Creative careers are changing fast at the tool layer and slowly at the judgement layer. There is a second thing fluency hides, too. Speed makes it cheap to keep going, and cheap iteration rewards people who never had to commit. Anyone trained where every frame cost money tends to reach a decision sooner and defend it longer. We are not nostalgic about that. We do notice it.
Constraint tolerance
Ramón Béjar's library runs on one rule: No Faces, Only Storytelling. Some people hear a rule like that as something taken away from them. The people we hire hear a decision already made, which frees them to spend everything on surface, architecture and light.
Christian Provenzano is the other version of the test. Restraint, a patterned-cap motif, a journal serif, packshots that stay clean and slightly withheld. A quiet brief is the hardest one to execute, because there is nothing to hide behind and no effect to reach for. Plenty of strong candidates can make something loud. Fewer can make something quiet and keep it interesting.
If the best answer to "why this frame" is that it looked good, we are the wrong studio.