From Idea to Image, Part Two

About this series

From Idea to Image is a weekly self-tutorial on making images with artificial intelligence. The text of this series is also written with the help of AI, and it reaches you edited and published by Owrang.

Each part teaches one skill and ends with a simple exercise you can do the same day. The three images at the foot of this page were made with exactly the steps you read below.

From last week's exercise, you now have two images. The second is closer to what you had in mind, and the reason is simple. The first image came from everything you said at once. The second came from one clear sentence about one specific change.

The whole secret of the prompt is right here.

A prompt is simply a description, and it works the way a description works between two people.

Imagine you are on the phone with a painter friend, explaining the picture you want. You would not read out a list of scattered words. You would say what is there, where it is, what light falls on it and how it should feel. The machine reads in much the same way.

So there is no formula to memorize. There is an order, and a few simple habits.

This week, step by step

Step 1

Write the short version first, just as it comes to mind.

Our example is this:

"An old bookshop with a feeling of longing."

Give this to the tool, make the image and keep it. This is image one.

Step 2

Now look at image one and ask yourself what the machine had to guess.

Which country
Which decade
Day or night
Whether anyone is in the picture
Where the light comes from
Where we are standing

Wherever the machine guessed, your picture and the picture it made moved apart.

Step 3

Now replace the general, emotional words with things that can actually be seen.

Longing, for example, cannot be drawn directly.
But these can be seen:

A glass of tea gone cold
A chair pulled a little back from the desk
Dust hanging in a beam of light

Feeling is good, but an image is built from visible things.

Step 4

Now write the description in order:
First the subject
Then place and time
Then light
Then mood
And last, the frame

Write in full sentences.

Our second prompt is this:

"A small old bookshop in Tehran in the 1970s, near sunset. Shelves reach the ceiling and books are stacked on the floor as well. An elderly bookseller sits behind a wooden desk, reading, with a glass of tea beside him. Warm light comes in through the front window and dust floats in it. The mood is quiet and touched with longing. Horizontal frame, seen from the doorway, like a photograph."

Now make the image. This is image two.

Step 5

Put the two images side by side.

Image one could be any bookshop.
Image two is yours.

What made the difference is plain detail that can be seen. Complicated or artistic words played no part.

Look at what the machine added on its own. A cat asleep on the carpet, a desk lamp, posters on the wall. A full description still leaves gaps, and the machine fills them with its own guesses. Keep the ones you like, and remove the rest with one sentence. The dust we asked for is barely there, so in the next round it moves to the first sentence.

Step 6

Now change just one thing, and say at the same time what should stay.

For example:

"Keep everything as it is. Move the viewpoint out to the street, so that we see the bookseller through the window glass at dusk, with the shop light as the only light."

This is image three.

Look closely at our image three. The machine kept the scene, the same desk, the same bookseller, even the cat on the carpet, and it switched on the desk lamp. But the sun is still shining at the end of the street, so the shop light is one of two lights. This is the moment for the next sentence. "Keep everything as it is. The sun has set and the sky is dark, so the shop lamp is the only light in the scene."

Step 7

Keep all three images, and save the words beside them.

Your prompts are part of your notebook. A month from now, if you want to reach the same mood again, you will need to know which description took you there.

When the machine goes its own way

Something you asked for is missing.
Move it into the first sentence and say it once, plainly.

Half the details have disappeared.
To begin with, five to seven visible details are enough. When the details pile up, the tool may pass over some of them or choose among them.

You asked for one change and everything changed.
Begin with "Keep everything as it is" and then name the one change.

Something keeps appearing that you would rather leave out.
Say so in one short sentence at the end, for example "The street is empty of people."

The image is close but the feeling is a little off.
Look at the light first. Most of the feeling in an image comes from its light.

A short working vocabulary

Most image tools understand these terms well. Knowing their English equivalents is useful too, because many common terms from photography and cinema are well known to image models.

Light
soft window light, backlight, hard midday sun, golden hour, a single lamp

Distance
close-up, medium shot, wide shot

Angle
eye level, from above, from below

Depth
a sharp subject with a soft background

Medium
photograph, watercolor, gouache, pencil drawing, linocut print

Next week we open each of these one by one.

The same order, another subject

The order you learned works for any subject. Here is a portrait:

"An elderly carpet weaver in a village workshop near Kashan, in the morning. She sits at the loom with her hands on the threads. Soft window light from the left falls on her face and hands. The mood is calm and concentrated. Vertical frame, medium shot, at eye level, like a documentary photograph."

Subject, place and time, light, mood, frame. The same order serves a poster, a book cover or an image for a poem.

Words inside the image

If your image needs words, a title on a poster or a name on a shop sign, leave them out for now. Writing is the least dependable part of these tools, and Persian script most of all. Sometimes the words come out right, as they did on the signs in our image two, and often they come out wrong, even when you give the machine the exact text. Look at those signs again. Nobody asked for them. The spelling happens to be right, but the sentences are the machine's own, and they are not ours. Writing inside an image has to be checked twice, once letter by letter and once for what it says. Make the image without text, and if the machine adds writing by itself, remove it with one sentence. In Part Eight we add the words ourselves, in a design program, exactly as we want them.

This week's exercise

Take last week's image.
Write its prompt again, this time in order:

Subject
Place and time
Light
Mood
Frame

Wherever you have used a general, emotional word, see whether you can trade it for something that can be seen.

Then make the image.
After that, change just one thing with one sentence and say what should stay as it is.

Keep all the versions.

How do I know I did it well?

Could someone who has never seen your image describe it roughly from your prompt alone?
Is every feeling word backed by at least one thing that can be seen?
When you changed one thing, did the rest stay?

For now, the aim is practice. You are learning to explain what is in your mind to the machine more clearly and more exactly.

Free tools for practice

The exercises in this series need no paid plan. Several free tools are available, and their allowances change often, so look at each tool's own page before you begin.

ChatGPT, free plan
It makes and edits images by conversation. The free allowance is small, so write and correct your prompt on paper first.

Gemini, free plan
It makes images within a daily allowance that Google adjusts from time to time.

Microsoft Copilot, free
It makes images from a written description. Speed and the number of images depend on how busy the service is.

Adobe Firefly, free plan
It offers a limited number of free generations that refresh every day.

One habit serves every free plan. Think on paper, finish the prompt, and then make the image. Treat every image as a real step in your work.

Next week

Frame, angle and light
Why can two images of the same subject feel so different?
The language of the image

The three images of this exercise

Version one: the short prompt
Version two: the full description
Version three: the same scene, seen from the street