Writing Good AI Prompts
Write clear prompts for AI templates: instructions or keywords, punctuation, weights, group photos and ready-to-use examples.
The prompt is the text that tells the AI what to make. A clear prompt gives consistent, controllable results, event after event. This guide covers the structure, punctuation and weights you can use in AI templates that take a text prompt, plus ready-to-use examples.
Chapters (8)
Chapters (12)
The basics
- English is the safest language for a prompt you type yourself. The questions for guests in a dynamic prompt can be in any language. Prompt Builder writes in the app's language by default, which most models follow well; if a model ignores parts of it, switch its Prompt Language to English.
- Keep it short: one or two lines is usually enough.
- Describe the scene, the clothes and the light, not the people. Then the same prompt works for one guest, a couple or a whole group.
- Ask the AI to keep the people, for example:
Keep every person's face, expression, pose and the number of people exactly the same. - Test on real photos: one guest, a couple and a group. Save prompts that work in the Prompt Library so you can reuse them.
Don't want to type? Since 5.8.0.0, Basic prompts in AI Online and AI Offline have Prompt Builder: pick the theme, outfit, lighting, mood, framing and more from lists, and it writes the prompt for you. It writes in the app's language by default; Prompt Language switches it. If a model follows one language badly, try another.
Instructions or keywords?
There are two ways to write a prompt, and both have their place:
| Style | Looks like | Use it when |
|---|---|---|
| Instruction | Change the clothes of the people in this photo to … Keep the same people … Do not add anyone. | The template changes the guest's own photo (an Image → Image model in AI Online, or Re-Imagine in AI Offline) and the guests must stay recognisable. This is the usual style for AI Online models. |
| Keywords | night street fashion, neon bokeh, rim light, 50mm lens, cinematic | You describe a look or a scene and want fine control with punctuation and weights (below). |
For keywords, a simple order works well: [Style/Theme], [Descriptive elements], [Techniques], separated by commas. Example: surreal landscape, swirling clouds, bold contrasting colors, impasto brush strokes, dramatic depth, crisp detail
Not every AI model reacts to punctuation and weights in the same way. They matter most for AI Offline looks; many online models just read the prompt as plain text. Always check the effect with your own template.
What each punctuation mark does
| Mark | Function | Example |
|---|---|---|
, comma | Soft separator between ideas. | mountains, river, blue sky |
; semicolon / . full stop | Hard separator between ideas. | ancient cavern; hidden treasure |
! exclamation mark | Emphasis. | cityscape at night! |
: colon | Sets an exact weight. | cat:2.0, eyes:1.8 |
( ) parentheses | Groups words and adds light emphasis. | (beautiful sunset) |
(( )), ((( ))) | Nested parentheses: stronger emphasis with each level. | ((detailed texture)) |
[ ] square brackets | Lowers the weight. | [sunset] |
| pipe | Blends two different ideas. | cyberpunk city | traditional batik |
Prompt weights
A weight tells the AI how important a word is.
| Value | Meaning |
|---|---|
1.0 | Normal (the default). |
above 1.0 | More important. |
below 1.0 | Less important. |
0 | Ignored. |
| Form | Effect |
|---|---|
keyword:value | Exact weight, for example cat:2.0. |
(keyword) | Light emphasis (about 1.1). |
((keyword)) | Medium emphasis (about 1.2–1.3). |
(((keyword))) | Strong emphasis (about 1.3–1.4). |
[keyword] | Less emphasis (about 0.9). |
You can combine them: a (cat:2.0), ((playful)), a [eyes:0.7], (curious)
Once you have weights that give good results, keep them and change only one value at a time while testing.
Group photos: keep men and women as they are
AI models tend to drift towards female faces in group photos. Use weights to keep men from turning into women (the exact numbers can vary with the rest of the prompt):
(man:1.6) or (woman:1.1)
If guests wear a hijab:
(man:1.6) or (woman:1.1) or (hijab woman:0.7)
- Make sure the number of people and their gender in the prompt match the photo you expect.
- Add
group shotfor group templates.
Keeping things out (no negative prompt)
Neither AI Online nor AI Offline Studio has a separate negative prompt box. To keep something out of the result, say so in the prompt itself, for example no brand logos or do not add anyone.
Prompt templates
Portrait, one person
portrait, [style], [face/skin/hair details], [wardrobe], [lighting], [lens], [background], [techniques], (detailed face:1.3)
Group photo
(man:1.6) or (woman:1.1), [style], [wardrobe], [lighting], [lens], group shot, [background], (detailed face:1.3)
Replace each part in square brackets with your own words, for example [lighting] → soft key light.
Ready-to-use examples
| Theme | Prompt |
|---|---|
| Surreal landscape | surreal landscape, swirling clouds, bold contrasting colors, impasto brush strokes, dramatic depth, crisp detail |
| Group, casual indoor | (man:1.6) or (woman:1.1), casual indoor portrait, soft key light, 35mm lens, mid-shot trio, neutral background, natural skin tones, subtle film grain |
| Night street fashion with hijab | (man:1.6) or (woman:1.1) or (hijab woman:0.7), night street fashion, neon bokeh, rim light, candid pose, 50mm lens, cinematic |
| Cyberpunk + batik blend | cyberpunk city alley | traditional batik patterns, neon glow, reflective puddles, moody atmosphere, volumetric light, ((texture detail:1.3)) |
| Close-up, focus on the eyes | portrait close-up, (detailed face:1.3), (sharp eyes:1.6), soft skin, studio lighting, 85mm lens, shallow DOF |
Checklist before you save
- A typed prompt is in English (the safest choice).
- It is short.
- The number of people and their gender are consistent.
- Weights you already tuned are left alone.
- You tested it on real photos: single, couple and group.
Save prompts that work in the Prompt Library so you can reuse them in other templates.
Next steps
Still need help?
Our team answers on WhatsApp. Tell us your Q-Booth version and what you see on screen.