Write structured prompts for Generative AI in Firefly
Jesús Ramirez shows how to art-direct results with detailed creative briefs for client-ready images.
Published
Introduction and meet Jesús Ramirez
[Jesús Ramirez]: My name is Jesús Ramirez, and I create TV and movie posters professionally.
In this video, I'll show you exactly how I prompt like an art director to get the professional results my clients demand.
Before we start, let's look at a few important concepts.
Generative AI is a technology that allows us to create images by describing them with words.
The two most important components for a designer are the AI model and the prompt.
The prompt is the words you use to ask for the image, and the AI model turns those words into the image.
Every AI model is trained differently.
Much like different photographers, they each have their strengths, styles, and way of communicating.
Think of your prompt as the creative brief you hand to that photographer.
Tell them exactly what you want.
The subject, mood, lighting, and everything your image needs, all in plain words.
Look at this example for a moment.
It's a mid-shot portrait composed on the rule of thirds, with strong Rembrandt lighting and leading lines pulling your eye right into the subject's head.
All within a complementary teal and orange color grade.
Everything I just described to you is something you as a creative professional have seen thousands of times.
But look at this.
This image is actually AI generated, and this was the prompt I used to make it.
All the elements I referenced a moment ago are there in the prompt.
This language is not new to you, you've been using it your entire career.
Let's look at how all this works in practice.
To start, go to the Firefly website at firefly.adobe.com and click on Create Image.
From here you can generate an image from scratch.
Learn how the model and the prompt work together
The first step is to select the AI model from the dropdown.
For this example, choose the latest Google Gemini model.
Gemini is great at following detailed plain language direction and keeping the same subject consistent across generations, which matters when you bring in your own content.
Then, set the options.
Every model has different ones.
In this case, you can set the aspect ratio and resolution that best works for your project.
Now you're ready to start crafting your prompt.
We'll start with a very simple prompt.
Just the word, "woman."
Then click Generate or press ENTER on Windows or CMD on the Mac to create your image.
And in a few moments, you'll see your result.
Here it is.
It's a woman, which is all we asked for.
The model decided everything else for us.
The pose, lighting, color grade, and location.
To get a result closer to what you have in mind, you have to ask for it.
This is where art directing comes in.
Let's add some direction and try again.
This time, type a full sentence.
"Woman in golden hour lighting, singing and dancing."
Then generate again.
And this is better.
We described the lighting conditions and her action.
But we still left many decisions to the model.
The lens, framing, color grade, and location.
Bring your own talent into the shot with a reference image
However, notice that no matter what the prompt is, it always invents a stranger.
In real client work, you often need to work with specific people or products, and those are virtually impossible to prompt.
Fortunately, you can use reference images in your generations.
To do so, click the image upload icon and choose an image from your device.
I'll upload this image of my talent to reference her in all future generations.
The reference loads as Image 1 and you can now call her directly in the prompt.
To save time, I'll paste in a new prompt.
To take real control, I've created a structured prompt formula that makes it easy to create and edit prompts.
It's like a complete written brief you will craft as an art director.
Think of this structured formula as a set of controls and ask yourself, what does your image require?
For this project, we need a subject.
Notice that the prompt references the photo we imported as Image 1, and it describes details about her outfit and hair.
Be as detailed as you like here.
I've also set her pose and the location.
Like any good brief from an art director, I've set the shot's composition, including the camera angle, her positioning in the frame, and the leading lines the street creates.
And of course, I gave directions for lighting, color grade, and even specifics about the camera lens I use.
Let's generate and wait for the results.
And here's the generated image.
Every control we set is there.
The location, pose, leading lines, and the autofocus background.
And most importantly, we've gone from a complete stranger to the talent our client requires.
To be clear, you can use a single paragraph using natural language to generate similar results.
The prompt formula is not a requirement.
However, the advantage of a structured prompt is that each piece can serve as its own control.
You simply find and edit the section you want to change without affecting the rest of the prompt.
For example, we can change the location to a rooftop bar in New York City.
Then, in composition, move her to the right third of the frame with a rounded bar creating leading lines towards her.
Leave every other element exactly as is and generate.
And here's the new version.
It looks fantastic.
Everything aligns with the new direction we took.
Speak the model's language with precise professional terms
Here's something else worth knowing.
Models like Gemini, Flux, and GPT carry a vast knowledge of the world.
That means you can reference precise professional terms by name and the model knows exactly what you mean.
For example, an exact Pantone color, a specific film stock a design movement, and even a real location.
I'll paste in a prompt and I'll generate.
In the meantime, let's analyze it.
It asks for an Art Deco editorial along Miami's Ocean Drive built on a Pantone 16-1546 Living Coral with its complementary teal and shot on a Kodak Ektar 100.
Every one of those is a precise name reference and the model rendered each one correctly.
This is your professional vocabulary and these models speak it fluently.
And don't worry if you don't generate the perfect image on the first try.
Just like a real photo shoot, the best shot is rarely your first.
Try different poses, compositions, and angles until it feels right.
And don't forget to experiment with different AI models.
You might find one that works better for the task at hand.
My advice is to treat each AI model like a photographer you're briefing.
Give clear direction, stay specific about what matters, and describe exactly what you want to see.
Now that you know how to art direct an image with a prompt, try it on your next project.
Again, my name is Jesús Ramirez.
Thank you for watching!
What you’ll learn
Learn how the model and the prompt work together
Generative AI turns words into images, and every model interprets them differently, like a photographer with their own style. Treat your prompt like a creative brief; describe the subject, mood, and lighting in language you already use.
Bring your own talent into the shot with a reference image
Upload a reference photo to keep a specific person consistent across generations. Use a structured formula to split the brief into controls, like subject, pose, and lighting, so you can edit one without touching the rest.
Speak the model's language with precise professional terms
Models recognize exact Pantone colors, film stocks, design movements, and real locations, so you can use them in prompts to dial in results. Expect a few rounds of testing poses, compositions, and even different models until the shot feels right.
Instruction by
