Write image prompts as structure, not adjectives
The point of the structure is that you can tell which part to change.
AI Tools Hub Editorial · published · updated
Five elements
| Element | What to state | Example |
|---|---|---|
| Subject | What is in frame | A notebook and pen on a wooden desk |
| Framing | Viewpoint and extent | Directly overhead, whole desk in frame |
| Light | Source and direction | Natural light from a window at left, soft shadows |
| Texture and colour | Overall register | Low saturation, warm, paper-like |
| Exclusions | What must not appear | People, text, saturated colour |
Drop the adjectives
"Beautiful", "professional" and "high quality" specify no direction, destabilise the result and obscure what actually worked. Replace them with conditions: generous whitespace, three colours maximum, level horizon.
Revise one slot at a time
- Too bright — change only light
- Too close — change only framing
- Colours too loud — change only texture and colour
Save what worked
Keep successful prompts with the image they produced, and record the model and version — the same words give a different picture elsewhere.
Length
Five elements fit in two or three lines. Longer than that usually means the same instruction stated twice; look for words you can delete without changing the output.
FAQ
Can I write in Japanese?
Depends on the service. Some accept it and still follow English more precisely. If the result misses, try English.