Five AI Image Prompts You Can Paste, and Why Each One Works

Every prompt guide teaches by fragment, ours included. Here is a weak line, here is a stronger one, here is the lesson. That is a good way to learn a rule and a poor way to get work done. When you actually sit down to make something, you do not want six illustrations of six principles. You want a whole prompt, finished, that you can paste and then push around.

So this post is a library rather than a lesson. Five complete prompts, a note on what order the parts belong in, and a short diagnostic for the render that comes back technically correct and somehow lifeless.

The order the parts go in

A prompt is read left to right and the front of it carries the most weight. The order that holds up: subject, place, light, shot, style. Subject first, because it is what the picture is of. Place second, because a real location decides half your lighting for free. Light third, named out loud, because it is the clause most people skip. Shot fourth: how much of the subject is in frame, and from where. Style last, because it recolours everything in front of it.

Thirty to sixty words is the range worth aiming at. Under thirty you have usually left out the light or the framing. Past sixty you are diluting: each extra clause competes with the others, and the attention the model can spend gets spread thinner across all of them.

Five prompts you can paste

Portrait, natural light.
“A woman in her early thirties sitting at a kitchen table by a window, morning light falling from the left across her face and shoulder, soft shadow on the wall behind her, waist-up, 85mm lens, shallow depth of field, warm neutral colour, photographic”

Note what is doing the work. Not “beautiful woman” but an age and a posture. Not “good lighting” but a direction, a time of day and what the light lands on. The shadow on the wall is there on purpose: it gives the room depth without you having to describe the room.

Full body in an environment.
“A man in a grey wool coat standing on a wet railway platform at dusk, overhead sodium lamps and their reflections in the puddles, full body, slight low angle, 35mm lens, cool blue shadows against warm lamp light, fine rain, photographic”

Full body is harder than a portrait, because the further back you stand the more of the frame the model has to invent. The fix is to spend words on the environment rather than the person. Two named light sources that disagree with each other, warm lamps against a blue dusk, will do more for the image than another sentence about the coat.

Product on a surface.
“A matte ceramic coffee cup on a dark walnut table, single soft window light from the right, long soft shadow falling left, one highlight along the rim, close three-quarter view, 50mm lens, shallow depth of field, dark neutral background”

Product shots live or die on the shadow. One light source, one direction, and an explicit instruction about where the shadow goes. “Studio lighting” on its own gets you a flat object floating in grey.

Landscape.
“A narrow alpine lake at first light, mist sitting on the surface, dark pine slopes on both sides, one snow ridge catching pink sun at the top, wide shot from the shoreline, deep focus, cool blues with a single warm accent, photographic”

Landscapes want scale and a time of day, and they want one thing to look at. The lit ridge is that thing. Without it you get a competent, forgettable postcard.

A start frame for video.
“A young woman with short dark hair leaning on a balcony railing at night, city lights out of focus behind her, warm light on her face from an open door to her left, waist-up, 50mm lens, she is still and looking off camera, photographic”

This one is written to be animated afterwards, which changes two things. The subject is deliberately at rest, because a still that already looks mid-action gives the video model nothing to move towards. And the background is out of focus, which hides the fact that backgrounds are where drift shows up first. Paste this into the Photo tab, then take the result over to Video and describe only the motion, as the video prompt guide lays out.

Style clauses replace vocabulary, they do not stack on it

The five prompts above all end in “photographic”, and swapping that one word out is the cheapest change you can make. But a style is not a filter you bolt on. It brings its own physics, and clauses from the old world start fighting it.

• Film noir:“high contrast black and white, hard key light through window blinds, deep black shadows, 1940s”
• Oil painting:“oil on canvas, visible brushwork, warm earth palette, canvas texture”
• Editorial fashion:“editorial fashion photograph, single hard strobe, clean seamless backdrop, high key”
• Physically based render:“3D render, soft global illumination, subsurface scattering on the skin, clean studio backdrop”
• Neon night:“neon-lit street at night, magenta and cyan practical lights, wet asphalt reflections, atmospheric haze”

Notice that “85mm lens, shallow depth of field” has no business in the oil painting prompt. A painting has no lens. Leaving it in does not make the painting cinematic, it makes the model hedge between two pictures. When you change the style, delete the clauses that belonged to the old one.

It rendered fine and it is boring anyway

This is the common case, and it is not a refusal or a failure. Something is missing rather than wrong, and each symptom points at a specific missing clause.

• Flat. No light direction was named. Say where it comes from and what it does to the shadows.
• Muddy colour. Too many hues competing. Name a palette of two or three and let one of them win.
• Plastic skin.The default beauty pass. Ask for what you want back: “visible skin texture, fine pores, unretouched”.
• Stiff pose. You described a state, not an action. Give the person something to do with their weight and their hands.
• Cluttered. No framing clause. Decide on close, waist-up, full body or wide, and say it.

Change one of these at a time. Two edits at once and you have learned nothing about which one helped, which is the whole reason keeping a character consistent is a discipline rather than a trick.

Write them in English

Your prompt goes to the renderer exactly as you typed it. Nothing translates it on the way, and these models were trained overwhelmingly on English captions, so English is the language in which they understand the most detail. Write the surrounding notes to yourself in whatever language you like, but keep the prompt itself in English and you will get a more faithful render for the same coins.

Two habits carry over from everything else we have written and are worth one line each: never name a thing you do not want, because naming it is how it gets drawn, and quality words like “8k masterpiece” are noise that buys nothing. Both are covered properly in the general prompt guide and, for adult scenes, in the 18+ one.

A Standard image is 40 coins, High quality is 60, and a generation that fails returns your coins automatically. Take any prompt above, paste it into Fellowi Images, change one clause, and run it again. That loop is the entire craft.