Images · The What2Prompt engine

AI Image Prompt Generator for ChatGPT, Gemini & Midjourney

A strong image brief chooses a subject, composition and purpose before choosing decorative adjectives. Use this AI image prompt generator for original photo concepts, editing instructions and illustrations in ChatGPT, Gemini or Midjourney. Keep the visual decisions portable, then adapt the syntax and reference controls to the tool you actually use.

See all prompt guides →
2Which AI will you paste it into?
Short and punchy.
1 credit per prompt

Works with ChatGPT, Claude, Gemini, Midjourney, Suno and more.

Copy · Paste · Done
STEP 3 · YOUR PROMPT
Prompt for MidjourneyQuickExample

Role

Act as an art director for an original educational image.

Context

A widescreen cover for a tutorial about miniature paper construction. I need one readable focal subject and space for a title added afterward. There are no real people or branded objects in the scene.

Task

One small paper greenhouse on a simple tabletop, folded cream panels and transparent window openings, one seedling visible inside, close three-quarter camera angle, soft side daylight, crisp silhouette against a muted mint background, gentle contact shadow, uncluttered right third for title space, tactile paper texture, restrained editorial still-life photography --ar 16:9

Format

Return the image line as written, followed by a short checklist for focal clarity, visible plant details and title space. Use only parameters available in the selected Midjourney mode.

Guardrails

No lettering, logos, extra buildings or named artist references. Do not add an unrequested version parameter. Inspect the final crop rather than treating the ratio as a resolution guarantee.
5 free prompts

Example prompts

Example

Midjourney: a single scene description

Role

Act as an art director for an original educational image.

Context

A widescreen cover for a tutorial about miniature paper construction. I need one readable focal subject and space for a title added afterward. There are no real people or branded objects in the scene.

Task

One small paper greenhouse on a simple tabletop, folded cream panels and transparent window openings, one seedling visible inside, close three-quarter camera angle, soft side daylight, crisp silhouette against a muted mint background, gentle contact shadow, uncluttered right third for title space, tactile paper texture, restrained editorial still-life photography --ar 16:9

Format

Return the image line as written, followed by a short checklist for focal clarity, visible plant details and title space. Use only parameters available in the selected Midjourney mode.

Guardrails

No lettering, logos, extra buildings or named artist references. Do not add an unrequested version parameter. Inspect the final crop rather than treating the ratio as a resolution guarantee.
Example

ChatGPT image: a protected photo edit

Role

Act as an editor preparing a GPT Image change request.

Context

I will attach my own photograph of an unbranded ceramic bowl. Preserve the bowl’s outline, glaze, camera angle and size. The purpose is an edited backdrop concept, not a new product design or factual packaging image.

Task

Replace only the cluttered background with a plain soft gray studio surface. Keep the bowl unchanged, retain the direction of the original light and use a believable contact shadow. Do not add food, labels, handles or decorations. Identify any uncertainty about the source image before changing protected details.

Format

Return one concise natural-language edit request and a review checklist: outline, glaze, light, shadow and unwanted additions.

Guardrails

No logos, invented product properties or real-person likenesses. The attachment defines the object; do not redesign it to fit a more fashionable composition.
Example

Gemini image: original anime-inspired illustration

Role

Act as an image-brief editor for a fictional garden story.

Context

The subject is an original small robot tending a seedling, not a licensed character. The image will appear beside a beginner gardening article. I want a gentle illustrated look rather than a real photo or imitation of a named studio.

Task

Create a soft watercolor anime-inspired illustration of a round garden robot kneeling beside one seedling. Use gentle linework, transparent pastel washes, a simple greenhouse background and soft morning light. Keep the two hands distinct and the watering can easy to recognize. Place the robot slightly left with quiet space on the right.

Format

Return one descriptive prompt and three checks for pose, object count and room for article text. Set any supported aspect-ratio control in the image tool itself.

Guardrails

No artist or studio names, real people, logos, licensed characters or invented lettering. Describe visual characteristics rather than borrowing a recognizable identity.

Choose the image’s job before its aesthetic

An image needs a purpose. A cover should communicate one idea at a small size; a storyboard frame should show spatial relationships; a product concept should preserve the real object’s features. Describe that job before the lighting or texture. “A ceramic bowl catalog concept on a clean backdrop” sets a useful constraint that “luxurious cinematic masterpiece” does not. Name the focal subject and which visual information the viewer must understand without reading a caption.

Next decide the composition. State close-up or wide view, subject placement and what should remain empty. If text will be added later, reserve space rather than asking the model to invent a slogan. A small thumbnail benefits from a clear silhouette and simple contrast, not a dozen background details. Review the image at its intended display size. You may need to remove elements that look attractive in a large preview but compete with the main message when reduced.

Keep portable decisions separate from tool syntax

At the time of writing, November 2026, image workflows include GPT Image in ChatGPT, Nano Banana models in Gemini and Midjourney. Their interfaces and modes need individual checks. A scene’s subject, light and composition can travel across tools; a parameter, reference control or editing setting may not. Store the portable brief separately from the tool-specific suffix or instruction. This makes switching tools a deliberate adaptation rather than a blind copy-and-paste experiment.

For Midjourney, a concise descriptive scene line with supported parameters is a useful pattern. For ChatGPT or Gemini edits, explain the requested change and the source image’s role in ordinary language. Do not imply that typing an output dimension guarantees that export size. Set supported controls in the selected product and inspect the resulting file. If an option is absent from your account, change the workflow instead of asking a prompt to create a feature.

Photo edits need a protected-details checklist

Editing differs from starting a new image. Specify the features you must not lose: an object’s outline, finish, position, scale and visible parts. Then list allowed changes. A background cleanup should not quietly recolor a product or change its angle. Keep the untouched reference available for comparison. Check contact shadows, edges and perspective as well as the object itself. A subtle inconsistency can make an otherwise polished image misleading or unusable.

Use real product photographs for factual claims and treat generated variations as concepts unless they faithfully represent the item. Do not let an AI add certifications, accessories, materials or packaging text. For personal or commercial material, confirm permission before uploading. An instruction to preserve a detail is a direction, not an enforcement mechanism; the tool can still alter it. Reject or repair outputs that change protected facts rather than lowering the review standard to match the result.

Describe style without borrowing an identity

You can express a visual style with line quality, shading, color relationships and medium. Soft watercolor anime-inspired art means transparent washes, gentle contours and expressive original characters; it does not require naming a studio or an artist. For editorial photography, describe lens feel, light direction and depth rather than asking for a particular creator’s signature. Keep practice subjects unbranded and wholly original, including fictional characters that do not copy a real person’s likeness.

If the output is wrong, identify the visible mismatch. Too many details need a simpler background; a flattened subject needs clearer light separation; a confusing action needs a different pose. Change one decision at a time and compare against the original purpose. Adding more style adjectives can make the instruction less consistent. Keep successful composition wording, but review every new result for object count, anatomy, accidental identity references and the truth of anything the image appears to claim.

What makes a good AI image prompt brief

  1. Name the subject and intended use before the style; a tutorial cover and a product evidence photo have different requirements.

  2. Specify framing, subject placement and protected details so the model is not left to design every decision.

  3. For photo edits, state what stays unchanged and request one bounded change to the supplied source.

  4. Describe visual styles generically and avoid real-person likenesses, brand marks or named studios and artists.

  5. Keep tool syntax separate from the portable image description; unsupported parameters can change or confuse the request.

Bad prompt vs What2Prompt prompt

Bad prompt

“Make an amazing cinematic anime image”

What2Prompt prompt

Role

Act as an art director for an original educational illustration.

Context

A small garden robot and one seedling, for a beginner gardening article.

Task

Soft watercolor anime-inspired style, gentle linework, pastel light, robot on the left with a clear watering can and quiet title space on the right.

Format

One readable widescreen composition with no lettering.

Guardrails

Use an original character; no named studios, artists, logos, extra hands or misleading product claims.

The improved brief makes subject, placement and purpose observable rather than relying on “amazing.” Its style description uses generic visual traits. You can carry those decisions between tools, but GPT Image editing, Gemini image workflows and Midjourney parameters are not interchangeable. Check the selected tool’s controls before pasting model-specific syntax.

Test a composition before spending effort refining texture. Inspect the subject silhouette, crop and object count at the final display size. For edits, compare the output with an untouched source; for product concepts, check every visible feature. What2Prompt writes the image brief, not the finished artwork. Generate in the chosen image tool, then review rights, accuracy and disclosure before publication.

Reference: Midjourney Prompt Basics ↗

Questions about AI image prompt

Can I use one image prompt in ChatGPT, Gemini and Midjourney?

Reuse subject, composition, light and protected details, then adapt the syntax. Midjourney examples may include supported parameters, while conversational image tools accept ordinary-language generation or edit requests. An uploaded reference needs instructions explaining its role. Do not assume every tool has the same editing, resolution or continuity controls.

How should I write photo editing prompts?

Identify a source photo you own or may process, list details to preserve and ask for one specific change. For example, replace a backdrop without changing an unbranded bowl’s shape or glaze. Check the result side by side with the source. A convincing edit can still introduce a false product feature.

Can I ask for an anime look without naming an artist?

Yes. Describe linework, shading, palette, facial proportions and mood generically, such as soft watercolor anime-inspired style. Use original subjects rather than recognizable characters or people. Avoid studio and artist names. Review the output for accidental resemblance, inappropriate details and the actual composition you requested.

What2Prompt prompt engine, with the W2P mark on cobalt.

Give your next idea a better brief.