From a few words to a full prompt: how to expand an idea step by step

Most image ideas start as a handful of words: “a cat in space”, “cozy reading nook”, “logo for a coffee shop”. Those words describe a subject, but an image needs dozens of other decisions before it exists. If the prompt doesn’t make them, the model makes them for you, and it picks the most common option every time.

Expanding a short idea into a full prompt is mostly a matter of noticing which decisions are still open and settling them on purpose. This guide lists those decisions, then expands three example ideas from one line to a finished prompt.

The decisions a short idea leaves out

Run any idea past these eight questions. Each unanswered one is a place where the result will be generic.

1. Subject detail

Not just what but which. “A cat” could be any cat. “A scruffy grey tabby with one torn ear” is a specific cat. Add one or two concrete details: age, material, texture, condition, clothing, pose, what the subject is doing.

2. Setting

Where is it, and what surrounds it? A setting gives the model a background, a scale and often a light source for free. “On a windowsill in a cluttered flat” does more work than “indoors”.

3. Composition

How is the subject placed in the frame? Close-up or wide? Centred or off to one side? Is there empty space for text? Composition is the decision people skip most often, and it is the one that most changes how a picture reads.

4. Camera and lens

For anything photographic, say where the camera is (eye level, low angle, overhead) and suggest a lens. “85mm portrait lens” implies a flattering, compressed close-up with a soft background; “24mm wide angle” implies space and context. Even for illustrations, a viewpoint helps.

5. Light

Name the source, direction and quality: “soft morning light from a window on the left”, “hard overhead sun”, “the glow of a laptop screen in a dark room”. Light sets the mood more than any adjective.

6. Palette

Two or three named colours steer the whole image. “Dusty pink and olive” and “cobalt and sodium orange” make the same scene feel completely different. Avoid “colorful” and “vibrant”; they don’t say which colours.

7. Medium and style

Photo, oil painting, ink drawing, 3D render, flat vector, film still? Choose one. Describe the qualities you want (“loose brushwork”, “clean geometric shapes”, “grainy film texture”) rather than naming a living artist.

8. Mood

One or two words for how it should feel: quiet, tense, playful, melancholy. Mood works best as a final note that confirms what the light and palette already suggest.

Three ideas, expanded

Idea 1: “a cat in space”

Start by answering the questions one at a time.

• Subject: a fluffy orange cat in a small, slightly oversized astronaut suit, helmet visor up
• Setting: floating inside the cramped cabin of a space station, cables and handrails around it
• Composition: medium shot, cat slightly left of centre, round porthole on the right showing Earth
• Camera: eye level with the cat
• Light: cool white cabin light from above, blue glow from Earth through the porthole
• Palette: white, steel grey, soft blue, with the cat’s orange as the only warm colour
• Medium: photographic, like a still from a science documentary
• Mood: curious, gently funny

Assembled:

A fluffy orange cat in a slightly oversized astronaut suit, helmet visor raised, floating inside the cramped cabin of a space station among cables and handrails. Medium shot at eye level, the cat left of centre, a round porthole on the right showing the blue curve of Earth. Cool white overhead cabin light and a soft blue glow from the window; steel grey and white surroundings so the cat's orange fur is the only warm colour. Photographic, documentary feel, curious and gently funny.

Idea 2: “cozy reading nook”

• Subject: a deep armchair with a knitted throw, a stack of worn paperbacks, a mug of tea on the armrest
• Setting: a window alcove in an old house, rain on the glass
• Composition: three-quarter view of the chair, window filling the right side of the frame
• Camera: slightly above seated eye level, 35mm
• Light: grey daylight from the window plus a warm reading lamp to the left
• Palette: oatmeal, rust and deep forest green
• Medium: interior photograph
• Mood: calm, sheltered

A deep armchair with a chunky oatmeal knitted throw in a window alcove of an old house, a stack of worn paperbacks on the floor and a mug of tea on the armrest. Rain streaks the window, which fills the right side of the frame. Three-quarter view from just above seated eye level, 35mm lens. Grey daylight from the window balanced by a warm brass reading lamp on the left. Palette of oatmeal, rust and deep forest green. Interior photography, calm and sheltered.

Idea 3: “logo for a coffee shop”

Logos need different decisions: there is no camera or light, but shape, simplicity and text matter far more.

• Subject: a coffee cup whose steam forms the outline of a mountain peak
• Composition: centred emblem, shop name underneath
• Text: the name in quotes, exactly as it should appear
• Palette: two colours only, espresso brown on cream
• Medium: flat vector, bold single-weight lines, no gradients
• Mood: friendly, outdoorsy

Flat vector logo for a coffee shop: a simple coffee cup whose rising steam forms the outline of a mountain peak, centred, with the name "Summit Roast" in rounded bold capitals underneath. Two colours only, espresso brown on a cream background. Bold single-weight lines, no gradients, no shadows, plenty of empty space around the emblem.

Formatting for your generator

The expanded description is the hard part; formatting is mechanical. Sentence-style prompts like the ones above suit Flux, Gemini and ChatGPT as written. For Midjourney, compress them into short phrases with the most important first and add parameters at the end, as the Midjourney parameters guide explains. For Stable Diffusion, convert them into tags and move unwanted elements into a negative prompt.

Our text to prompt tool does both steps: type a short idea and it fills in the missing decisions, then formats the result for the generator you pick. Treat its output the way you would treat the expansions above, as a draft you can adjust one decision at a time. Browsing finished examples in the prompt gallery is also a quick way to see how other people answered the same eight questions.

Common mistakes

• Adding quality words instead of decisions. “Masterpiece, ultra-detailed, 8k” doesn’t settle anything. Replace them with light, lens and palette.
• Stacking styles. “Watercolour, photorealistic, anime” averages into mush. One medium, at most one modifier.
• Contradictions. “Minimalist” with a long list of objects, or “dark and moody” with “bright sunny day”.
• Burying the subject. Put the main subject in the first sentence. Many models weight early words more heavily.
• Negations in the main prompt. “No people” can introduce people, because the word is there. Use your generator’s negative prompt or exclusion option where it has one, or describe the scene positively: “an empty street”.
• Changing everything between runs. If the result is close, change one decision and regenerate, so you know what made the difference.

The same ideas from the angle of individual habits are covered in seven prompt habits that fix generic results, and model-specific wording for ChatGPT is in our DALL-E prompt tips.

Checklist

Before you generate, make sure your prompt answers:

1. Which specific subject, doing what?
2. Where, and what surrounds it?
3. How is it framed?
4. Where is the camera, and which lens?
5. Where does the light come from, and what is it like?
6. Which two or three colours?
7. Which single medium or style?
8. What mood?

Then format it for your model, generate, and change one thing at a time.