Every AI image generator is a genie that grants exactly what you asked for — which is the problem, because most people don’t know what they asked for. “A beautiful landscape” is not a prompt; it’s a shrug with extra steps. The generator fills every unspecified dimension with its defaults, and the defaults are engineered to be inoffensive, which is to say, boring.

Prompt building is the discipline of specifying the dimensions that matter. Our AI Image Prompt Builder turns it into a fill-in-the-blank process — subject, style, lighting, camera, mood — producing a copy-ready prompt for any generator. Free, no signup. This guide teaches the underlying structure so you can build great prompts with or without the tool.

The six blocks of a complete prompt

Think of a prompt as six slots. You don’t need all six every time, but you should know which ones you’re leaving blank — because blank slots get filled by defaults.

Block 1: Subject (the anchor)

The most important block, and the one people under-specify most. “A woman” is not a subject; it’s a category. A subject has concrete, visualizable specifics:

  • Weak: “a coffee cup”
  • Better: “a handmade ceramic coffee cup”
  • Strong: “a handmade speckled ceramic coffee cup, steam rising, on a oak saucer”

Rule of thumb: if a police sketch artist couldn’t draw it from your description, it’s not specific enough. Materials, colors, condition, and one distinguishing detail are the minimum.

Block 2: Style and medium

This block sets the entire visual language. Name it explicitly:

  • Photography: “editorial fashion photography,” “documentary street photography,” “macro product photography”
  • Illustration: “flat vector illustration,” “watercolor illustration,” “ukiyo-e woodblock print style”
  • 3D: “octane render,” “pixar-style 3D character,” “isometric 3D illustration”
  • Era/movement: “1970s film photography,” “art deco poster,” “renaissance oil painting”

One named style beats five adjectives. “Cinematic” is doing real work; “beautiful, amazing, stunning, gorgeous” is not.

Block 3: Lighting

Lighting is the highest-leverage block most beginners skip entirely. It determines mood, depth and realism more than any other technical choice:

  • Direction: “soft window light from the left,” “backlit at sunset,” “overhead studio lighting”
  • Quality: “soft diffused light” (flattering, commercial) vs “hard dramatic light” (edgy, high-contrast)
  • Color: “warm golden hour tones,” “cool blue moonlight,” “neon pink and teal”
  • Special: “rim lighting,” “candlelit,” “volumetric god rays”

If your images look flat and generic, the missing ingredient is almost always a lighting specification.

Block 4: Camera and composition

Borrow the vocabulary of photography even when generating illustrations — generators understand it:

  • Angle: “low angle,” “overhead flat-lay,” “dutch angle,” “eye-level portrait”
  • Framing: “extreme close-up,” “wide establishing shot,” “rule of thirds composition”
  • Lens/depth: “shallow depth of field,” “85mm portrait lens look,” “everything in sharp focus”
  • Aspect: match your intended use — portrait for phones, landscape for banners (and set --ar in Midjourney or the equivalent)

Block 5: Mood and atmosphere

The emotional register, in one or two words: “melancholic,” “energetic,” “serene,” “luxurious,” “eerie.” Mood words steer color grading and micro-decisions the other blocks don’t cover. Don’t stack more than three — they start contradicting each other.

Block 6: Negatives (what to exclude)

Most generators support negative prompts (Midjourney’s --no, Stable Diffusion’s negative field). Use them for the artifacts you keep seeing: “no text, no watermark, no blurry, no extra fingers, no distorted faces.” Negatives are especially valuable for product and portrait work where generators’ typical failure modes are predictable.

The builder’s formula: three worked examples

Example 1 — product shot:

“A matte black wireless headphone on a reflective dark surface, dramatic product photography, single spotlight from above, deep shadows, dark charcoal background with subtle blue rim light, luxurious, ultra-detailed, no text, no watermark”

Blocks: subject (matte black wireless headphone + surface) ✓, style (dramatic product photography) ✓, lighting (spotlight, rim light, deep shadows) ✓, camera (implied close product framing) ✓, mood (luxurious) ✓, negatives ✓.

Example 2 — portrait:

“Portrait of an elderly fisherman, weathered face, wool sweater, documentary photography style, soft overcast daylight, shallow depth of field, muted blue-grey harbor background, contemplative mood, 85mm lens look”

Example 3 — illustration:

“Isometric 3D illustration of a cozy coffee shop interior, warm lighting, tiny detailed furniture, soft shadows, pastel color palette, cheerful, clean minimal style, no text”

Notice each prompt is 30–50 words. Completeness comes from covering the blocks, not from length.

The iteration method: change one block at a time

When a result is wrong, beginners rewrite the whole prompt. Don’t. Diagnose which block failed and change only that:

  • Wrong object/scene? Subject block — add specifics.
  • Right subject, wrong vibe? Style or mood block.
  • Flat, boring, generic? Lighting block (this is the fix 70% of the time).
  • Weird framing or crop? Camera block — add angle and aspect ratio.
  • Recurring artifacts? Negatives block.

Keep a small log of prompt → result → change. After a dozen iterations you’ll have personal knowledge of which phrases your generator of choice actually respects — because generators differ, and no guide can substitute for that.

Adapting one prompt across generators

The six blocks are universal; the syntax isn’t:

  • Midjourney: append parameters — --ar 16:9 for aspect, --no text, watermark for negatives, --style raw for less beautification. Midjourney responds well to comma-separated descriptive phrases.
  • DALL-E: prefers natural sentences over keyword strings. Rewrite the blocks as flowing description: “A matte black wireless headphone sits on a reflective dark surface, lit by a single overhead spotlight…”
  • Stable Diffusion / Flux: use the negative prompt field for block 6, and consider adding quality tags the community uses (“masterpiece, best quality”) — these are model-specific conventions, not universal truths.
  • Firefly / Ideogram: strong at text rendering; if your image needs words in it, say so explicitly and keep the text short.

Our Image to Prompt tool is useful in reverse here: generate an image, analyze what worked, and fold the successful phrasing back into your builder templates.

The five prompt sins

  1. Adjective stuffing. “Beautiful amazing stunning gorgeous incredible” — the model averages them into mush. Pick the two strongest descriptors and delete the rest.
  2. Contradictory blocks. “Photorealistic” + “watercolor illustration” in the same prompt. The model picks one; you don’t control which.
  3. Prompting the UI, not the image. “4k, 8k, trending on artstation, unreal engine” — these are cargo-cult tokens from 2022 prompt forums. Some still nudge quality; most just add noise. Describe the image you want instead.
  4. Ignoring aspect ratio. A portrait-composition prompt rendered at 16:9 will crop or stretch your intent. Set the frame first.
  5. No negatives for known failure modes. If every portrait comes back with garbled hands, “no deformed hands” in negatives is not optional — it’s the fix.

The bottom line

Good prompts are specifications, not wishes: a concrete subject, a named style, deliberate lighting, camera language, one mood, and explicit negatives. Build them block by block — or let our AI Image Prompt Builder assemble the structure while you supply the ideas. And when you need proven starting points for product work specifically, our product photography prompt swipe file has 100+ copy-paste recipes organized by lighting, angle and style.