Understanding direction in film is essential, whether you’re directing other people, or receiving it yourself.

So this week, let’s break down the key - maybe the key concept that I think helps understand how to direct or work with directors. I’d say it’s even relevant if you’re directing AI.

The agony and the ecstasy at the heart of the creative process is that the goal isn’t known before you start. However sure you feel about the end result, the one thing that you can be sure of is that when you get there it’s going to be different to what you’d imagined. Maybe a little, maybe a lot. Life in general is the same - do any of us have the lives our childhood selves envisioned? The journey changes the goal.

This means that good direction shouldn’t be goal based, because you’re asking someone to do something that you yourself can’t define or locate. The extreme of this is the note that strikes fear into the heart of every professional working in film - “I’ll know it when I see it”.

Worst. Direction. Ever.

Let’s break this down into giving and receiving direction, although they’re two sides of the same coin:

Giving Direction

For any thing that you’re creating for a film, there are an infinite number of things that it could be, but only one thing that it should be. “I’ll know it when I see it” implies that you should leap feet first into the infinite, and keep hauling out glittery cosmic findings, until you find the one.

A much better approach is to accept that you don’t know the goal, but you know where it’s found. A vector - that’s a direction plus a magnitude. How far should you go, and in what direction?

If you’ve worked with the person before and you have built in trust, you can let them go pretty far, if you’re confident of the direction. If not, just a short way. Then they report back what they’ve found - showing you the work they’ve done. It generally won’t be right yet, in which case it may have negatives, and positives. You’ll notice both, but the positives are better than the negatives, because we already know that there are infinite things that a thing can’t be. Then you direct them again. Fine tune their direction, let them know how far to go. Repeat until they find the thing.

Directing actor and puppet on set of Constant Parasite

Receiving Direction

Your objective is to find your vector at all costs. Many directors will tend towards goals. “You should design a creature that has four horns, four eyes, and two tails”. That might be useful information, if the script needs those specific attributes. But if it does, then they should just give you the script.

I’d argue it’s not direction, because it has no vector. “When I see this creature, I need to feel at a glance like it’s threatening, but as I look closer I can see there’s wisdom there”. Ok, that’s a direction. Now the magnitude - how fast or how far should you go? You should suggest as small a distance as possible, because then you can easily assess if the direction is wrong. This is huge. Some doodles, a moodboard, AI, whatever it takes. Just a few steps - is this the direction? If it is, you can go further faster after, until you find the thing

Directing AI

I remember a couple of years ago, when photoreal AI imagery was brand new to the public, a friend of mine prompted Midjourney to make “the most beautiful person in the world”. He assumed that having scraped billions of faces, it would be able to create a face that was objectively the best one.

Of course, it created someone generically good looking. Just an average of faces that exist in the overlap of “beautiful” and “man/woman” in latent space. That’s an AI problem - it averages, so it doesn’t do well at extremes - but more importantly for this discussion, it’s a direction problem.

When directing humans, using their perspective - their humanity - is useful for many. For some (the auteurs), it’s their vision. When directing AI, you’re an auteur, because it has no humanity. Don’t ask for beauty, tell it what beauty means, otherwise you’re going to get an average.

Long prompts mean greater distances that the AI travels before checking in, and it’s travelling through a multi-dimensional land humans can never understand or enter. So try shorter ones, and build up - that way you can get more data on its journey. When editing video or image, always keep the prompts short so that you can control its journey.

By the way, when it comes to directing an acting performance in AI, you actually want to do it in a totally different way to how you direct an actor on set!

You never want to tell an actor HOW to act. But with AI, you don't want to give it big beats "Be scared, be determined" because it's not going to interpret that in the same way that a human would. So you want to give it very specific performance beats, e.g "Blink. Then swallow hard. then open your lips to speak, but change your mind and close them again, firmly"