Learning vocabulary for AI image prompts means learning the visual words that tell an image generator what should appear, where objects should sit, and how a scene should feel. Start with subject, composition, framing, viewpoint, foreground, background, soft light, contrast, focal length and aspect ratio. These terms are useful for illustrations, social-media graphics, product images, invitations, concept art and photographs.
A good prompt usually needs fewer clever adjectives and more concrete relationships. Instead of “make a beautiful café picture,” try “a cup of black coffee on a wooden table in a quiet café, seen from slightly above; warm morning light from the left; open space on the right for a title; landscape composition.” The second prompt describes objects, viewpoint, light and layout. It gives a creator—or an AI model—a visible result to aim for.
This guide explains how to learn and use the vocabulary, not merely how to press a generation button. OpenAI Academy recommends starting with the purpose, subject, action, setting and intended visual style, adding composition and lighting only when they matter. Models can still make errors, especially with precise text, brands, anatomy or very crowded scenes. Always inspect outputs rather than assuming a descriptive prompt guarantees an exact result.
Quick answer: seven parts of a useful image prompt
- Purpose: what is the image for—a thumbnail, invitation, article illustration, product mock-up or diagram?
- Subject: what must the viewer see first? Name the main person, object or activity.
- Setting: what background, location and period are essential?
- Composition: where should each important element appear? Use left, right, centre, foreground and background.
- Viewpoint: is the camera eye-level, above, below, close or far away?
- Lighting and mood: should the scene have soft daylight, hard shadows, diffused studio light or warm sunset colours?
- Constraints: what must remain unchanged and what should be absent—extra text, stray objects, unwanted logos or cropped feet?
Contents
- Essential visual vocabulary
- Camera and framing words
- Lighting and colour words
- Four prompt clinics
- Editing without accidental changes
- Four-week practice plan
- Frequently asked questions
Definition lock: the visual language behind an image
| Term | Plain-English meaning | Example in a prompt |
|---|---|---|
| Subject | The principal object or person the image is about. | A red umbrella held by a commuter. |
| Setting | The surrounding place and circumstances. | A rain-soaked urban walkway. |
| Composition | How the important visual elements are arranged. | Place the umbrella in the left third. |
| Framing | How much of the subject and scene fit inside the picture. | Waist-up portrait; both hands visible. |
| Viewpoint | Where the scene is seen from. | At eye level, slightly to the right. |
| Perspective | How depth and size relationships appear from a viewpoint. | A long corridor receding into the distance. |
| Foreground | The part of the scene nearest the viewer. | A blurred plant along the front edge. |
| Background | The setting farther from the camera. | A row of distant shopfronts. |
| Negative space | Relatively uncluttered area around the subject. | Keep the right third open for a headline. |
| Focal point | The part of the image intended to attract attention. | A single brightly coloured notebook. |
| Aspect ratio | Relationship between image width and height. | 16:9 landscape or 9:16 portrait. |
| Visual hierarchy | The order in which a viewer notices elements. | Headline first, main subject second, supporting text third. |
Composition is not the same as style. A watercolour image can have a centered composition or an off-centre composition. A photorealistic image can use either shallow or deep focus. The vocabulary becomes useful when you separate the subject, arrangement, lighting and medium. Changing one of these does not automatically require changing all the others.
Camera and framing vocabulary: see the scene before describing it
| Term | What it looks like | When it helps |
|---|---|---|
| Wide shot | Shows a broad setting and often the entire subject. | Establish where an event happens. |
| Medium shot | Shows a person or object at moderate distance. | Balance facial expression and surroundings. |
| Close-up | Shows a small part prominently. | Emphasise an object, expression or material. |
| Extreme close-up | Shows a tiny detail. | Show fabric texture, a droplet or a dial. |
| Eye-level | Viewpoint approximately level with the subject’s eyes. | A familiar, relatively neutral perspective. |
| Low angle | Camera looks upward at the subject. | Emphasise apparent height or presence. |
| High angle | Camera looks downward at the subject. | Reveal a layout or make the subject seem smaller. |
| Top-down / overhead | Camera looks almost straight down. | Lay out food, tools or documents. |
| Three-quarter view | Subject seen partly from the front and partly from the side. | Reveal shape and depth. |
| Shallow depth of field | A smaller distance range appears sharply focused. | Separate a subject from a softened background. |
| Deep depth of field | More of the scene is sharply focused. | Show a landscape or several objects clearly. |
| Focal length | A camera-lens measure associated with framing and perspective choices. | Specify a wide-angle or portrait-lens feel. |
| Crop | Part of a composition is excluded by the image boundaries. | Ensure an object is fully included when necessary. |
A prompt saying “portrait, close-up, full body, top-down” asks for conflicting framings. Choose the primary requirement. If the shoes must be visible, say full body visible, including feet, and avoid also requesting an extreme close-up of the face. If the image is intended for a tall phone screen, state the portrait aspect ratio separately from the camera viewpoint. Portrait orientation is not the same as a portrait photograph.
Did you know? A focal length is a cue, not a guarantee of physical optics
Real cameras use lenses with measurable focal lengths, but many image generators interpret “50 mm portrait lens” as a visual style cue. It does not prove the resulting geometry follows exact lens physics. When accuracy matters, describe the visible result directly: natural perspective, eye-level medium close-up, moderate background blur. Then inspect the composition. This is more reliable than stacking camera brands and numbers without explaining their purpose.
Lighting and colour words that change the atmosphere
| Term | Simple meaning | Visible effect |
|---|---|---|
| Soft light | Light with gradual shadow boundaries. | Gentle facial shading and subdued contrast. |
| Hard light | Light that creates sharper shadows. | Strong shape definition and contrast. |
| Diffused light | Light spread or softened through a broad source. | More even illumination across the scene. |
| Backlighting | Light originates behind the subject. | A glowing rim or silhouette, depending on exposure. |
| Side lighting | Light comes mainly from one side. | Reveals form and surface texture. |
| Rim light | A highlight near a subject’s edge. | Separates the silhouette from its background. |
| High-key | A mostly bright image with relatively light tones. | Airy, clean or optimistic appearance. |
| Low-key | A predominantly dark image with selective highlights. | Intimate, dramatic or mysterious appearance. |
| Warm colour temperature | Light with visually yellow-orange tones. | Sunrise, lamps or warm interiors. |
| Cool colour temperature | Light with visually blue tones. | Overcast or deliberately cool styling. |
| Golden hour | Warm, low-angle light near sunrise or sunset. | Longer shadows and warm ambience. |
| Contrast | Difference between light and dark or between competing visual elements. | Directs attention and affects readability. |
| Saturation | Intensity of colour. | Controls whether colours feel muted or vivid. |
| Monochrome | A visual based on one colour family or greyscale. | A restrained, unified look. |
Be careful with the word natural. It does not tell a generator whether the scene should be brightly lit or shadowy. “Natural daylight from a large window on the left, with gentle shadows” is more usable. Likewise, “cinematic” is too broad unless you specify the visible elements you want—framing, light, colour and atmosphere.
Four prompt clinics: make vague requests precise
Clinic 1: a realistic everyday photograph
Before: “Beautiful photograph of a student reading.”
After: “A fictional teenager reading a paperback at a wooden library table, natural posture, medium shot at eye level, soft daylight through windows on the left, book and hands clearly visible, calm realistic colour, no visible school logos or readable private documents.” The second version defines who is present, what the person is doing and how the scene should be framed.
Clinic 2: a vertical social-media illustration
Before: “Make a poster about vocabulary.”
After: “Portrait 9:16 educational poster, illustrated open book in the lower centre, five small icons showing speech, reading, listening, writing and memory, clean light background, large clear negative space in the upper third for a short headline, limited colour palette, simple shapes, no paragraphs of tiny text.” The improved prompt separates image layout from words that might require manual typesetting.
Clinic 3: a product photograph
Before: “Professional bag photo.”
After: “Studio product photograph of a generic tan canvas tote bag standing upright on a light grey seamless background, three-quarter front view, handles fully visible, diffused light from upper left, natural contact shadow, no brand marks or printed text, ample space around all edges.” The important words are upright, three-quarter, fully visible, diffused and contact shadow.
Clinic 4: a scene showing depth
Before: “A beautiful park with a path.”
After: “Wide landscape image of a quiet public park, gently curving path entering at lower left and leading toward a small pavilion in the background, trees framing the sides, soft early-morning light, foreground grass with natural texture, realistic proportions, no people.” The phrase leading toward identifies a visual relationship, not merely an object list.
How to write spatial instructions without confusing the model
| Phrase | What it means | Useful prompt fragment |
|---|---|---|
| Left of | An object appears to the viewer’s left of another. | The notebook is left of the mug. |
| Behind | An object appears farther away and partly obscured, where appropriate. | A bookshelf stands behind the desk. |
| In the foreground | Closer to the viewer. | Small flowers appear in the foreground. |
| In the background | Farther from the viewer. | Mountains rise in the distant background. |
| Centered | Near the middle of the frame. | Place the logo-free box at the centre. |
| Along the edge | Near a defined boundary. | A subtle border follows the lower edge. |
| Fully visible | No essential part should be cropped. | Keep all four wheels fully visible. |
| Empty space | Intentionally uncluttered area. | Leave the top quarter free for later lettering. |
A prompt with ten objects but no relationships often gives an unpredictable layout. A prompt with only four objects but clear positions can be much easier to review. For a teaching infographic, name the order of sections and use short labels rather than asking an image generator to invent a dense page of prose. For exact numbers, spellings and factual diagrams, verify the final rendering and use ordinary design software for corrections when needed.
Editing an existing image without accidental changes
A new image prompt asks a model to invent a scene. An edit prompt describes what must change in an existing scene. If the task is small, state the change first and then describe what must remain unchanged. For example: “Change only the notebook cover from blue to green. Preserve the notebook size, desk, light direction, hand position, crop and every other object.” This is the difference between redesign and local revision.
| Editing verb | What it asks for | Precision check |
|---|---|---|
| Replace | Swap one named object or attribute. | What replaces it, and where? |
| Remove | Delete a specified element. | Must the background be reconstructed? |
| Preserve | Keep specified features unchanged. | Which features are non-negotiable? |
| Reframe | Change how much of the scene fits. | Will any essential part be cropped? |
| Recolour | Change selected hues or palette. | Which areas must retain their original colours? |
| Relight | Alter illumination while keeping scene geometry. | What direction and softness should the light have? |
| Composite | Combine elements from different visual references. | Which source supplies each object, scene or style? |
| Iterate | Refine the previous result through a new change. | Did only the intended dimension improve? |
For image references, name their different roles: “Image 1 is the subject; Image 2 is the lighting reference. Retain the subject’s appearance and pose from Image 1 while adapting the soft window lighting from Image 2.” Even good image models can drift. If an edit moves the subject or changes the face unexpectedly, restate the preservation requirements and revise one change at a time.
The five mistakes that cause weak image prompts
- Conflicting framing: asking for an extreme close-up and a full-body shot simultaneously.
- Unmeasurable praise: saying “make it amazing” without describing a visible feature.
- Unspecified layout: listing objects without left, right, centre, depth or relative size.
- Overloading camera jargon: listing brands and focal lengths without explaining the visual goal.
- Skipping review: assuming generated text, fingers, labels, logos, object counts or real-world facts are perfect without inspection.
Four-week practice plan: from naming objects to controlling a composition
| Week | Practice theme | Repeatable task | Success check |
|---|---|---|---|
| 1 | Subject and setting | Write five one-sentence prompts, each with one clear subject and setting. | Can a friend identify the intended focal point? |
| 2 | Composition and viewpoint | Make three variants of one scene: eye-level, top-down and low-angle. | Does viewpoint genuinely change while the subject remains? |
| 3 | Light and medium | Describe the same scene in soft window light, hard studio light and warm evening light. | Can you explain the visual difference without saying “better”? |
| 4 | Editing and quality control | Refine one result through four single-change edits with explicit preservation constraints. | Are changes local, legible and appropriate to the purpose? |
Use active recall rather than rereading definitions. Close the glossary and describe a scene using six terms from memory. Then compare the resulting image against the specification. Interleave different scenes—a person, room, product and landscape—so the words become transferable rather than tied to a single template. The goal is world first, word later: notice the visible effect, then learn the precise name for it.
Frequently asked questions
Do I need to know photography to make a useful AI image prompt?
No. Everyday descriptions of subject, position, lighting and purpose already help. Photography vocabulary is valuable when you need more reliable control over a scene’s appearance.
Does a longer prompt always create a better image?
No. More words can introduce contradictions or distract from the important constraints. Start with the visible essentials, inspect the result, then refine one feature at a time.
Does saying “8K” guarantee a real 8K image?
No. A descriptive phrase in the prompt does not override the output dimensions supported by a particular tool. Check the actual file size and generated resolution.
What is the difference between 9:16 and a portrait photograph?
9:16 is a tall aspect ratio. A portrait photograph is a kind of subject or genre, usually emphasising a person. A landscape scene can be made in a 9:16 canvas, and a person’s portrait can be horizontal.
How do I make text in generated images more reliable?
Use short exact text, quote it, identify its position and style, and check the final image for spelling and legibility. For dense educational content or formal documents, consider adding text in a design editor afterwards.
How should I use a real person’s photograph as a reference?
Use references with permission, describe exactly which visual features should be preserved and avoid presenting synthetic changes as authentic documentary evidence. Always respect privacy and consent.
Official sources and related eduKateSG learning
- OpenAI Academy — Creating Images with ChatGPT: practical prompt structure, images and text.
- OpenAI — Image Prompting Guide: subject, composition, constraints and controlled edits.
- Adobe — Leading Lines Photography: a compositional principle used in both photography and prompt language.
- eduKateSG — Learning Vocabulary for ChatGPT Prompts: foundational instruction verbs.
- eduKateSG — Top 20 AI Vocabulary Words: general AI concepts.
Next useful step: Find one image you like. Describe its subject, viewpoint, foreground, background, lighting, colour and aspect ratio in your own words. Then change one term and predict how the image should change. You will have learned vocabulary by connecting it to something you can see.