Every prompt in this gallery is someone's actual working file — the text that produced the image, copied straight from the creator's public post. At just over 1,100 prompts (890 English, 134 Chinese, 86 Japanese at the time of counting), that's a corpus big enough to ask a simple question: what words do people actually use when writing image prompts? So we counted. Every word, every prompt, one afternoon.
Prompts get described as incantations — mystical wordings that summon an image. Counted word by word, they read more like design briefs.
The top of the list is layout vocabulary
The single most common content word across the whole English corpus is "white" — 999 occurrences, with words often repeating within a prompt — followed by "background" (847), "black" (801), "small" (792), "text" (631) and "composition" (563). Position words are everywhere: "left" (671), "right" (665), "top" (464). "Vertical" appears 463 times.
White background, left, right, top, small, vertical. These are the words of someone laying out a page, not someone describing a dream. The Chinese side of the corpus has the identical skeleton: 背景 (background, 92 uses), 构图 (composition, 64), 主体 (subject, 62), 留白 (negative space, 42), 竖版 (vertical format, 36). Different language, same brief.
A strong prompt answers layout questions — what's in frame, what the background is doing, where things sit, what the aspect ratio is — before it reaches for adjectives. You can watch it in this split-poster prompt, where the whole composition is specified as coordinates: upper half, lower half, mapping rules.
Color words carry the style
After the layout layer, color does the heavy lifting: "white" (999), "black" (801), "blue" (541), "red" (397), "warm" (607), "dark" (558). That white/black dominance is not a coincidence — it's the visual language of the current era of AI images: product shots, stickers, posters and characters composed on clean grounds. If you want your prompt to look contemporary, the data says lead with the ground color.
The hedging layer
Then comes a layer you'd never guess from reading prompt guides: the hedges. "Natural" (700), "soft" (697), "subtle" (477), "slightly" (475), plus "visible" (364) as its inverse. These words do quality control — they tell the model to stop short of the extreme. The Chinese corpus hedges the same way: 自然 (natural, 74), 真实 (realistic, 67), 保留 (keep/preserve, 55), 少量 (a small amount, 50). "Slightly" is doing more work in modern prompts than any poetic flourish.
And the art-direction register
The most striking high-frequency word of all is one nobody romanticizes: "editorial" (219 uses) — more common than "style" itself. Along with "minimalist" (118) and "poster" (362), it shows prompts reaching for a professional register, the language of magazines and galleries. The corpus isn't asking for fantasy; it's art-directing.
What to do with this
- Write your prompt like a brief: subject, background treatment, layout, format — then style. The order matters more than the poetry.
- Use position words deliberately (left/right/top, three-quarters, centered). 670+ uses of "left" say they work.
- Pick a ground color early. White is the current default for a reason.
- Hedge instead of exaggerating: "slightly", "subtle", "natural" keeps the model from overdoing an effect. Exaggeration words like "extremely" barely appear.
- Repetition is the idiom. The corpus repeats its vocabulary heavily — you are not being boring by reusing "natural light"; you are speaking the language.
The full method, for anyone who wants to reproduce it: strip stop words, count stems across the raw prompt text, and read the top hundred. The list you get is this list — a parts catalog for the images of the current moment. And the pages behind the pieces mentioned here show the same vocabulary at work on individual prompts, one editorial note at a time.
