Composing a scene: foreground, subject, background

A village's events committee wants an illustration for the summer banquet poster. The initial prompt is one line: "outdoor village banquet with people laughing, a long table, paper lanterns, a church in the background." The image contains all of those elements, stacked at the same level, all the same size and all equally sharp. It looks like an inventory rather than a scene.

An image is read in depth. The eye enters through the foreground, stops on the subject, then explores the background. If you don't describe those three levels, the tool places them at random.

The three planes

The foreground is whatever is closest to you. It frames the image and gives a sense of scale: branches, a glass sitting on a tablecloth, a hand serving food. It's often slightly blurred, which sets the subject apart.

The subject is the element the eye should reach first. Preferably just one. Say where it is (in the center, on the left third) and what it's doing.

The background places the scene: the setting, the depth, sometimes a distant landmark like a bell tower. It's calmer and softer than the rest, otherwise it competes with the subject.

For the banquet, the prompt becomes:

illustration of a summer village banquet at dusk, in the foreground a few blurred paper lanterns and a glass of wine on a checkered tablecloth, in the middle ground a long wooden table with a group of neighbors laughing around a roast, the table slightly left of center, in the background a small stone church and a row of plane trees fading into soft evening haze, warm string lights overhead, gouache style, calm empty sky on the right for a title

The subject is the group around the table: it's named, placed, and busy doing something. The foreground blurs and frames. The background fades into haze, which puts it in its place.

What helps the tool respect your planes

• Use words of distance. "In the foreground," "in the middle," "in the distance," "fading into haze."
• Describe the depth of field. A blurred foreground and a sharp subject tell the tool that attention goes to the middle. The photo vocabulary gives terms like "shallow depth of field" or "medium shot."
• Don't go beyond three or four elements per plane. Every object you add dilutes the others. A table, some guests, and a light are enough.
• Keep a single light source. If you mention both the setting sun and the party's neon lights, the image hesitates. One dominant light, and at most one secondary.
• Leave empty space if the image is going to carry a title. It's better to plan for it from the composition: text is easier to add by hand than to generate.

When the planes blur together

The common mistakes are easy to recognize. The subject disappears because you named four of them. The background is as sharp as the foreground, and the image looks flattened. A background element grows and moves in front. A few answers:

1. Keep only one subject, then add the other elements one by one in new generations.
2. Specify position ("on the left," "in the distance") rather than listing objects.
3. Add "blurred background" or "faded backdrop" if the setting is too present.
4. If a small element stays wrong, fix it locally with inpainting.

To change one variable at a time and understand what's having an effect, the logic of iterating on a prompt works for images too.

An exercise on an image you know

Pick a real scene you could describe from memory, such as the terrace of a café you know. Write three lines: what you see in the foreground, who or what occupies the middle, what fades into the background. Put them together in a single prompt, ending with the light and the palette, for example with the help of the article on colors and mood. Generate four images and note the one where the eye immediately knows where to go.