Image generation basics
Subject, style, and constraints
A useful image prompt names what must appear, how it should feel, and what the finished asset must be able to do.
1Try it yourself
Playground
Image prompt builder
Subject + style suffix — the same pattern as creative-tools, step by step.
Before you start
Why this matters
Start with a brief, not a spell
Image generation can feel like guessing a secret phrase. It is more reliable to treat prompting as writing a compact creative brief. The model is not reading your intention behind “make it cool.” It is translating words and other inputs into visual patterns. Your job is to make the important choices visible in the request.
A strong starting prompt has three layers:
- Subject: the people, objects, setting, action, and important relationships in the scene.
- Style: the visual medium, mood, palette, texture, lighting, and degree of realism.
- Constraints: the format, audience, placement needs, required details, and boundaries the result must respect.
Consider “a robot in a garden.” It establishes a subject but leaves nearly every production decision open. A more useful brief is:
A small weathered gardening robot kneels beside raised vegetable beds, gently tying a tomato vine to a bamboo stake. Friendly editorial illustration, cut-paper texture, muted greens and warm coral accents, soft morning light. Landscape 16:9 hero image with calm negative space on the left for a headline; no text or brand marks.
This prompt does not guarantee a perfect image. It gives you observable criteria for deciding whether an output is close.
2Learn the idea
Read
Define the subject precisely
Subject detail should resolve ambiguity, not bury the model in adjectives. Name the subject’s defining attributes and the action that matters. “A scientist” may produce a generic lab coat. “A marine biologist in orange rain gear labeling water samples on a small research boat” establishes identity through visible context.
Useful subject questions include:
- How many primary subjects are there?
- What are they doing rather than merely being?
- Which objects establish the setting or story?
- What relationship must be clear between subjects?
- Which attributes are essential, and which are optional?
Countable requirements deserve explicit language: “one bicycle,” “three ceramic cups,” or “a single person.” Models can still miscount, so inspect rather than assume. If the image carries an instructional claim—such as correct safety equipment—verify every visible detail.
Read
Describe style with visual properties
Style is broader than naming an art movement or creator. It includes medium, rendering, lighting, color, texture, era, and mood. “Cinematic” alone is underspecified. “Low-key studio lighting, shallow depth of field, cool shadows, restrained amber highlights, subtle 35 mm grain” gives concrete signals.
Try combining properties from several categories:
- Medium: documentary photograph, ink drawing, clay miniature, vector illustration, watercolor.
- Rendering: flat shapes, delicate cross-hatching, realistic materials, simplified geometry.
- Palette: monochrome navy, earthy neutrals, high-contrast primaries, pastel gradient.
- Light: overcast daylight, hard noon sun, rim lighting, diffused window light.
- Mood: energetic, contemplative, playful, clinical, mysterious.
Avoid piling on conflicting styles. “Minimal flat vector, photorealistic oil painting, rough pencil sketch” asks for incompatible visual systems. Pick a dominant treatment and add only details that support it.
Read
Add production constraints early
Constraints are not an afterthought. A beautiful square portrait may be useless for a wide website banner. State the intended use and the properties required by the layout:
- Aspect ratio or orientation.
- Space reserved for copy, controls, or cropping.
- Minimum legibility at thumbnail size.
- Background requirements, such as plain, transparent, or environmental.
- Colors that must harmonize with an interface.
- Elements that must not appear.
Treat generated typography cautiously. Some systems render short text better than others, but exact spelling, accessibility, and localization remain fragile. A dependable workflow often generates the visual without text and adds approved typography in a design tool.
Read
Prioritize the brief
Longer is not always better. When every detail is emphasized, nothing is prioritized. Put the central subject and action first. Follow with composition, then style, then technical and exclusion constraints. Repeat only a requirement that is genuinely easy to miss.
Separate hard requirements from preferences:
- Must: one red backpack, empty upper-right corner, no visible logos.
- Prefer: late-afternoon warmth, subtle film grain, candid mood.
This separation helps when evaluating tradeoffs. An image that misses a preferred texture may still be usable; one that places the headline area behind a face is not.
Read
Build your first prompt
Choose an asset you could actually use: a presentation cover, a product concept, a classroom diagram, or an editorial illustration. Write one sentence for each layer:
Subject: Who or what is shown, where, and doing what?
Style: Which medium, palette, light, texture, and mood fit the audience?
Constraints: What dimensions, layout space, required objects, and exclusions make it usable?
Combine the sentences, generate several candidates, and score them against the same brief. Variation is expected. The prompt is a specification for exploration, not a deterministic rendering command.