How to Write AI Image Prompts That Actually Work
By The Fellowi Team · · 6 min read

Most disappointing AI images do not come from a weak model - they come from a prompt that never actually said anything specific. “A beautiful woman in a fantasy setting” is not a description, it is a category. The model has to guess everything you did not say, and it guesses the most average, generic version it can. The fix is not a magic keyword or a longer list of adjectives - it is naming the specific things you already have in your head and just have not written down yet.
Below are six real before-and-after rewrites. The pattern repeats every time: swap a generic noun for a specific one, add a real place instead of an abstract mood, and describe the shot - lens, light, framing - the way you would describe a photo you actually took.
1. Name the subject, not the category
Before: “a fantasy warrior”
After: “a battle-worn elven ranger with silver braided hair, a scar across one eyebrow, wearing weathered leather armor with a green cloak”
“A fantasy warrior” could be a thousand different images, so the model picks the most statistically average one. The moment you give a specific hair color, a specific scar, a specific armor material, there is only one image left to converge on.
2. Give it a real place, not a mood word
Before: “a cozy scene”
After: “a small reading nook by a rain-streaked window, a steaming mug on the sill, an open book face-down on a worn armchair”
“Cozy” is a feeling, not a place. Describe the actual objects and the actual room, and the feeling shows up on its own - you do not need to ask for it twice.
3. Describe the shot, not just the subject
Before: “a portrait of a woman”
After: “a 35mm portrait, shallow depth of field, soft window light from the left, subject looking just past the camera, natural skin texture”
Think like a photographer for one sentence: what lens, what light, what is the subject doing with their eyes. These are the details that separate a flat, stock-photo-looking result from one that feels like a real, specific moment.
4. Product shots need a surface and a light source
Before: “a nice product photo of a perfume bottle”
After: “a frosted glass perfume bottle on a wet black slate surface, single soft key light from the top right, dark background, subtle reflection below the bottle”
“Nice” does not mean anything to the model. A surface, a light direction, and a background choice are three decisions a real product photographer makes before pressing the shutter - make them yourself instead of leaving them blank.
5. Landscapes and concept art want scale and time of day
Before: “an epic landscape”
After: “a lone lighthouse on a cliff at dusk, storm clouds breaking to reveal orange light on the horizon, waves crashing far below, wide-angle composition”
“Epic” is doing no work here. Time of day, weather, and a sense of scale (a small lighthouse against a huge sky) do the job that “epic” was trying and failing to do.
6. Original characters need consistency anchors
Before: “an anime girl character”
After: “an original anime character: short teal hair with a side braid, heterochromia (one amber eye, one blue), wearing a cropped denim jacket over a white tank top, standing on a city rooftop at golden hour”
If you plan to generate the same character again later, write down the two or three details that make them recognizable - hair, eyes, a signature piece of clothing - and repeat those exact words every time. Consistency comes from repeating your own anchors, not from the model remembering who they were last time.
Quick do’s and don’ts
- Name the specific thing, not the generic category it belongs to
- One clear mood or lighting choice beats five stacked adjectives
- Say what you want in the frame, not what you do not want
- Describe the shot (lens, light, angle) the way you would describe a real photo
- Keep a short list of your own anchor details for any character you want to repeat
Ready to try one of these? Open Fellowi Images and write your prompt - your first Standard image is already covered by the free coins on a new account. Once your prompt is right, quality and format are the next decision - see when Standard is enough and when High quality is worth it.
Keep reading

Three Image Models, One Studio: Which One to Pick for Which Picture
The Model select in the image studio has three entries, and the cheapest one is deliberately limited. Here is what each produces, what each costs in coins, and when the small one is the right call.

Focus and Backgrounds in Adult AI Character Art: Notes on Depth of Field
Eight lab notes, each a before and after prompt line, on getting the right thing sharp and keeping a loud background from pulling the eye off the character.

AI Concept Art for an Indie Game: One Week from Mood to Key Art
Monday moods, Tuesday environments, Wednesday a character who keeps his look, Thursday key art. Plus what stays outside it: no sprite sheets, no tiles, no engine-ready assets.