Prompting Like a Creative Director

When I first started working with image generators, I approached prompting the way a lot of people do: describe the image, add some camera terminology, specify the lighting, and see what comes back.

That works to a point.

But after spending more time with tools like Midjourney, I realized that the most useful thing I bring to the process isn't knowing how to write a complicated prompt. It's the same thing I've brought to photography and creative direction for years: knowing what I'm trying to make and being able to recognize when I'm getting closer to it.

Good prompting, at least for me, starts before the prompt itself.

Start with the idea

Before a shoot, I wouldn't begin by choosing an 85mm lens and then try to figure out what the photograph should be about.

The concept comes first.

The same applies to generative imagery. Before I start describing lighting, lenses, or color, I want to know what the image is supposed to communicate.

Is it energetic or quiet? Aspirational or uncomfortable? Polished or imperfect? Is the subject the most important part of the frame, or is the environment doing most of the storytelling?

Those decisions give everything that follows a reason to be there.

A technically detailed prompt can still produce a meaningless image if there isn't an idea underneath it.

Build a visual hierarchy

I tend to construct prompts the same way I think through a photograph: subject, environment, light, composition, mood, and then the smaller details.

Not every element deserves equal weight.

If the location is essential to the concept, I'll establish it early. If a particular expression or gesture matters more, I'll put more attention there. I usually start with less information than I think I need and add detail as the direction becomes clearer.

It's tempting to keep adding adjectives when an image isn't working. I've found that removing information can be just as useful.

Sometimes the model isn't missing a description. It's being asked to do too many things at once.

Photography knowledge helps, but only when it has a purpose

My photography background naturally finds its way into prompts.

I'll reference focal length, depth of field, direction of light, camera position, framing, film characteristics, or the qualities of a particular type of lens when those things help describe what I'm seeing in my head.

But technical terminology isn't a shortcut to a good image.

An 85mm lens isn't inherently better than a 35mm lens. Rembrandt lighting isn't more sophisticated than flat daylight. Those choices matter because they change how an image feels.

The question is always why.

If I want intimacy, I might bring the camera closer. If the environment matters, I might use a wider perspective. If I want something to feel less produced, perfect studio lighting may be exactly the wrong choice.

That's no different from working with a real camera.

Don't expect the first generation to be the answer

I rarely think of prompting as a single instruction.

It's closer to directing a shoot.

You start somewhere, look at what you're getting, identify what's working, and make adjustments.

Maybe the composition is right but the lighting feels too commercial. Maybe the wardrobe works but the location doesn't. Maybe there's something unexpected in the image that's more interesting than the original idea.

When that happens, I follow it.

I'll change one or two things rather than rewriting everything. That makes it easier to understand what's actually affecting the result.

The image itself becomes part of the conversation.

Learn to edit, not just generate

One of the strange things about generative AI is how easy it is to make something visually impressive.

That can also be a problem.

When almost every generation has some level of polish, selecting the right image becomes more important.

I ask many of the same questions I'd ask while reviewing photographs from a shoot:

Does this actually communicate the idea?

Does the subject feel believable?

Is there anything distracting from the image?

Does it fit with the rest of the campaign or project?

Would I still like this if I didn't know it was made with AI?

That last one is especially useful.

Novelty can make an image seem better than it is.

Develop your own visual language

I save prompts, references, lighting ideas, and approaches that produce interesting results, but I don't think of them as formulas.

They're more like a working vocabulary.

Midjourney's reference and moodboard tools can also be useful here. Rather than chasing whatever visual style is currently popular, I can use my own work and references to push generations toward something that feels closer to my taste.

Over time, patterns start to emerge.

That's more interesting to me than finding a perfect prompt.

Anyone can copy a prompt. Taste is harder to copy.

Know when prompting is finished

Generative tools are only one part of my process.

There is a point where generating another fifty variations isn't making the work better. Once I've found the direction, I'd rather move into editing, retouching, compositing, motion, or whatever else the project needs.

The goal isn't to create an image entirely inside one tool.

The goal is to make the image.

That's ultimately what prompting like a creative director means to me.

You're not trying to prove how sophisticated your prompt is. You're making decisions, giving direction, responding to what you see, and knowing when something is ready to move forward.

The technology keeps changing.

Those skills don't.