The Audition

Everyone building with AI video right now is fighting the same fight. They want the face to stay the same from one shot to the next. Consistency. It's all over the threads, the tools are built around it, and the worry never quite goes away, because the face slips the second you look away from it.

I think it's the wrong fight to pick first.

Before you worry about whether a face will stay the same, there's a harder question worth asking. Can this face actually act. Can it hold a real performance and still come back as itself on the other side. That one comes first, and almost nobody asks it, because the tools never made them. In film you ask it on the first day. You audition before you cast. You bring a face in, you push it a little, and you watch what happens when it's under some pressure. The point was never to look at the face. You already saw the headshot. The point is to find out whether the person is still there once you ask them to feel something.

So that's what I did. Not a consistency test. An audition.

One invented face, and then the plain things you'd ask of any actor reading for you. Sit neutral. Give me a real smile. Now grief. Now anger. Now fear. Turn to profile. Nothing else moves the whole time. Same light, same distance, same plain wall behind her. The only thing allowed to change is what the face is doing. That part matters, because if you let the light and the lens move too, you can hide almost anything. Hold it all still and the face has nowhere left to hide.

Then you just look.

Top row is one woman. Neutral, then joy, then grief, then anger, then fear, and then she turns away. She's the same person in every single frame. The cheekbones, the set of the eyes, the freckles across the nose, all of it stays put while the feeling moves through her. You believe it's one actor doing six things, because it is.

Bottom row got the exact same asks. And it comes back a different man every time.

Here's the part I want you to sit with, because most people have it backwards. The bottom row isn't bad at feeling. Look at him. He smiles, he cries, those are real tears, he's genuinely scared in the one where his eyes go wide. The emotion is all there. He's just not the same person twice. The grief is one man and the fear is another and the smile is a third. Range with nobody home. That's not a character. That's a casting call that never ended.

And that's the trap nobody really warns you about, because everyone's busy staring at the consistency problem instead. People picture the failure as a face going blurry or wrong. It usually isn't. It looks like this. Six convincing, deeply felt, completely different people. Every frame is fine on its own. Line them up and the character is just gone.

So why does the top face hold when the bottom one comes apart.

It's the same thing a good casting director could tell you in about five seconds. Structure. The top face has landmarks. The cheekbones, the strong line of the nose, the freckles sitting in one specific pattern. When a big expression yanks everything around, those are the things the face can come home to. She can twist into a snarl or fold into grief and still resolve back to the same woman, because the bones underneath are loud enough to read straight through the emotion.

The bottom face has none of that to hold onto. Smooth, even, easy to like, the kind of face that looks great in a single neutral frame and wins you over in the headshot. But there's nothing to return to. No anchor. So the moment it has to do something, it drifts, and the next strong expression tugs it somewhere else again, and you end up with six strangers.

This is old knowledge, honestly. It's why casting was never just about who photographs well in a still. The faces that work on screen are the ones that read, the ones with something specific in them, because those are the faces that survive being directed. The flawless average face is the one that quietly lets you down the second it has to carry a scene. Directors figured this out a hundred years ago. The generation crowd is about to learn it again, one drifting edit at a time.

Here's the practical version, and it barely costs you anything.

Audition the face before you lock it. Before you build a whole project on a character, run them through the range. Make them smile, make them grieve, make them rage, make them turn away. Lay the results in a row and look. If the same person keeps coming back, you've got a character, and your consistency problem just got a lot smaller, because the face wants to stay itself. If six people come back instead, at least you found out now. Not after you've cut twenty scenes around a face that was never going to hold. Now, while it still only costs you an afternoon.

Consistency is downstream of casting. That's really the whole point. Everyone's trying to fix at the end what they should've settled at the start. They lock a face because it looked good in one frame, then spend the rest of the project fighting to hold something that was never built to hold. The answer isn't a better consistency tool. It's a better audition.

Cast a face that can act, and most of the consistency takes care of itself. Cast one that can't, and nothing downstream is going to save it.

You can always tell the work that auditioned. The face is doing the work, it's the same face doing it, and after a moment you stop thinking about the face at all and just watch the person. That's the whole game. The second the audience starts thinking about the face, you've lost them. A face that holds is one they forget to question.

So before the next character, bring the face in. Ask them to feel something. See who walks back out.