The five parts of a prompt that actually work
Subject, setting, light, framing and finish — the checklist behind every good result.
SamImageGenerator ·
Most prompts that disappoint have the same shape: a subject and nothing else. A prompt that works usually carries five things, and you can write them as one plain sentence rather than a list of keywords.
The five parts
| Part | What it answers | Example |
|---|---|---|
| Subject | What is in the picture | a dusty 1970s station wagon |
| Setting | Where it is | on a gravel road through pine forest |
| Light | What time and what mood | low golden hour sun from the left |
| Framing | How much you see | wide shot, camera at eye level |
| Finish | How it looks as an image | shallow depth of field, warm film grain |
Strung together that is one sentence, not five: "a dusty 1970s station wagon on a gravel road through pine forest, low golden hour sun from the left, wide shot at eye level, shallow depth of field and warm film grain."
You do not need all five every time
Padding a prompt to reach a word count makes it worse, not better. When a result feels generic, the fastest fix is to check which of the five you left out. Nine times out of ten it is light or framing, because those are the two people forget they are allowed to ask for.
Light is the single biggest lever on whether an image reads as a photograph or an illustration. "Overcast", "one lamp on the left", "backlit at dusk" each change the whole picture. Framing is the second: "close-up" and "wide shot" of the same subject are two different images, and the model will pick one for you if you do not.
Where the style and the ratio fit in
The finish overlaps with the style picker, and that is fine — pick the style closest to what you want, then spend your words on what the style cannot say. The framing overlaps with the aspect ratio, which is why the ratio is chosen before generating rather than cropped afterwards. Write the sentence, pick the ratio, generate four variants, keep the closest one and edit from there.