Back to Journal
Prompting-9 min read

Prompting Is a Vocabulary Problem60 words that sharpen your AI prompts

AI gives you exactly what you can name — nothing more. The words below are the difference between “make it better” and getting what you meant.

The vocabulary that sharpens AI prompts

Every guide to prompting teaches the same techniques: give examples, ask for steps, assign a role. They work, and you can learn all of them in an afternoon — which is why they are not where prompts actually fail.

Prompts fail at the adjective. A model asked for a better logo, a nicer email, or a more professional tone returns the statistical average of everything ever called better, nicer, or professional — the beige at the center of its training data. The model isn't being lazy. The prompt is. Ask instead for a backlit, desaturated portrait with shallow depth of field, or an apology that is earnest and plainspoken with no hedging, and the output snaps toward your intent.

Wittgenstein wrote that the limits of language are the limits of one's world. In the age of prompting that stopped being philosophy and became an operating manual: your vocabulary is the resolution at which you can specify what you want. Here are sixty words that raise it — each linked to a full illustrated entry you can actually learn.

Light

Light is the first thing an image model decides and the last thing most prompts mention. One of these words does more than a paragraph of adjectives.

golden hourthe warm, low light just after sunrise or before sunsetbacklitlit from behind, so subjects rim with lightsilhouettea dark shape against light; detail surrendered to outlinechiaroscurobold contrast of light and shadowtenebrismchiaroscuro pushed to drama: subjects spotlit out of blacksfumatoedges melted into smoke — no lines, only transitions

Color & surface

Palette words set mood faster than mood words do. "Sad" gets you clichés; desaturated gets you winter.

saturatedcolors at full intensitydesaturatedcolors drained toward gray — memory, winter, griefmonochromeone hue and its shadessepiathe warm brown of antique photographspastelpale, chalky, low-saturation coloropalescenta milky play of shifting colorsmatteflat and non-reflectiveglossysmooth and mirror-shiny

Composition & lens

Where the camera stands and what it keeps sharp is a vocabulary, not a vibe. These are the words photographers already use, and models already know.

rule of thirdssubjects on the gridlines, a breath off-centernegative spaceemptiness used as a compositional elementdepth of fieldhow much of the scene is in focus, near to farbokehthe soft glow of out-of-focus background lightsvignetteedges darkened to pull the eye inwardwide-anglemore scene, stretched perspectivebird's-eye viewlooking straight down from aboveisometric3-D without perspective, like a game diorama

Tone & register

For text, the highest-leverage words describe voice. "Professional but friendly" is a shrug; these are specifications.

wrydry, twisted humor with a straight facedrollamusing in an odd, understated waydeadpanhumor delivered without any signal that it's humortongue-in-cheekmeant ironically, not literallysardonicscornfully mocking; grimly cynicalmordantwit that bites and corrodesacerbicsharp and sour in toneirreverentcheerfully disrespectful of the sacredearnestserious and sincere, zero ironyclinicaldetached, precise, emotionlessmatter-of-factunemotional even about dramatic thingsmeasuredcareful, unhurried, deliberateunderstatedsaying less than you could, on purposeplainspokendirect and unadornedcolloquialthe register of everyday speechbreezycasual, light, effortlessly informalfolksyhomespun, neighborly informalitylyricalsonglike, expressive, emotionally richelegiacmournful for something lostoraculargrand, cryptic, propheticclippedbrisk, cut-short deliverytelegraphicstripped to essential wordspurple prosewhat to ask for less of: ornament that upstages meaning

Technique

The craft terms themselves — the shared language of people who prompt for a living. Know them and the documentation, the papers, and the error messages all start making sense.

system promptthe standing instruction a model reads before your messagepersona promptingassigning the model a role to shape its answersfew-shot learningteaching by example, inside the promptchain-of-thoughtasking for reasoning steps before the answerself-consistencysampling several reasoning paths and taking the votetree-of-thoughtbranching reasoning that can backtrackcontext engineeringcurating everything the model sees, not just the questiondelimitermarkers that separate instructions from datastructured outputforcing the answer into JSON or another schemanegative prompttelling an image model what to leave outaspect ratiothe frame's proportions — 1:1 postcard or 21:9 cinemastyle transfersame content, different styletemperaturethe randomness dial: low for precision, high for surpriseprompt injectionhostile input that tries to override your instructionsjailbreaka prompt crafted to slip past a model's rules

None of these words is exotic. That is the point: they sit just past the edge of most people's active vocabulary, in the gap between recognizing a word and reaching for it. Prompting rewards the reach. The photographer's terms, the editor's tone words, the engineer's craft terms — they were all precision instruments before models existed. Models just made them executable.

Make the reach automatic

Segue's Prompting collections — Light, Color & Composition and Tone & Register — turn these words into flashcards, quizzes, and typing practice, so they're in your fingers the next time the cursor blinks.

Open the Prompting collections