How to use these prompts
The basic mistake is describing only the subject and hoping. The AI fills the gaps however it happens: light, angle and style it chooses itself. The more items you fix, the more the image resembles the one in your head. The sequence that works is always the same: first the subject, then how it's shot, finally the atmosphere.
A note on register: a single honest mood word beats three enthusiastic adjectives. "Melancholic" guides better than "beautiful, moving and unforgettable", which tells the model nothing.
The prompt library
The base scheme, to fill in
Create an image with these elements:
- Subject: [what or who, with a detail that characterizes it]
- Style: [realistic photo / illustration / watercolor / 3D / comic]
- Framing: [close-up / wide shot / from above / at eye level]
- Light: [golden sunset light / soft light / backlight / neon]
- Atmosphere: [a single word: serene, dramatic, cheerful, melancholic]
- Format: [square / vertical / horizontal]
Consistent portrait of a character
Create the portrait of this character, described so that it stays
recognizable if I regenerate it: age, build, hair, clothing,
expression. Style [realistic photo / illustration]. Simple background
that doesn't distract. Soft light on the face.
Character:
Image for a social media post
Create an image for a post on this theme, designed to stop the scroll:
clear subject in the center, empty space at the top where I'll insert
the text, colors that stand out on a small screen. Vertical format.
Post theme and brand tone:
Variations starting from one that works
This description generated an image I like. Give me three variants
changing one thing at a time: one with different light, one with
different framing, one with different style. Keep the subject fixed.
Description that worked:
Improve a prompt that gives poor results
This prompt gives me disappointing images. Rewrite it adding the items
that are missing (style, framing, light, atmosphere) and removing the
vague words. Explain in two lines what you changed and why.
Current prompt:
A concrete example
Paolo runs a farm stay and wants an image for the page's cover. He starts badly: he writes "photo of the countryside" and gets an anonymous catalog landscape. He switches to the base scheme and fills it in: subject "table set outdoors with bread and wine, farmhouse in the background"; style "realistic photo"; framing "at eye level, table in the foreground"; light "golden late-afternoon light"; atmosphere "welcoming"; format "horizontal".
The result is already usable, but the farmhouse is too far off. Instead of rewriting everything, he uses the variations prompt changing only the framing: he asks for a version "slightly wider, farmhouse more visible". He keeps the rest fixed. On the third generation he has the cover. The difference wasn't a better AI, it was a prompt that told it what to do.
When it does NOT work (and how to fix it)
If the image ignores a detail you asked for
Generators sometimes skip an element, especially if the prompt is crowded. Move the important detail to the beginning and make it essential: "a red bike (central element, must be present)". What you put at the start and mark as central carries more weight.
If faces or hands come out badly
It's a known weak point of generators. For faces, ask for less close-up or three-quarter framing; for hands, avoid complex poses or hide them (in pockets, behind the back). If you need a perfect face, a real photo remains the best choice: the AI is for illustration, not for the ID document.
If you can't find where to enter the prompt or the format options
Generator interfaces change often, and buttons move. If you can't find the format option, write it inside the prompt ("vertical 9:16 image"). If you can't find where to upload a reference image, ask the AI in the text: "use this description as a style reference". The text box is the command that never disappears.
A tip from someone who actually uses it
Change one variable at a time. When an image is almost right, the temptation is to rewrite the whole prompt: that way you also lose the things that were working and start over from scratch. Modify only the light, or only the framing, and compare. Build your image through small corrections, the way you focus a lens, not through fresh restarts.
Frequently asked questions
Can I use generated images for my commercial activity?
It depends on the tool: each has its own terms of use, and they change. Before using an image to sell or advertise, check the terms of the generator you used. Some grant commercial use, others limit it in the free plans. The fix is to read the conditions once, not to guess.
Why are two images from the same prompt different?
Because there's a random part in the generation: the same prompt never gives two identical images. It's an advantage when you're looking for variants, an annoyance when you want to repeat a result. To get closer to repeatability, fix as many details as possible in the prompt and, where the tool allows it, reuse the same starting point that gave you the good result.
Does a longer, more detailed prompt always give a better image?
No, beyond a certain point it reverses. A prompt overloaded with adjectives confuses the model, which starts ignoring items at random. Better five precise items (subject, style, framing, light, atmosphere) than twenty adjectives. Clarity beats quantity: every extra word must add an information, not an enthusiasm.