Sketch / 12 words
Find the subject
“A handmade ceramic teapot on a dark wooden kitchen counter.”
This AI image prompt is enough to test the central object, but composition and atmosphere remain open.
Creative brief field guide
AI Image Prompt Guide for Clear, Controllable Results
An AI image prompt works better as a small creative brief than a bag of impressive adjectives. This guide shows how to describe visible decisions, control relationships, and revise a weak result without rewriting the whole scene. Each line of the AI image prompt should change something a viewer can point to.
Prompt specimen6 visible decisions
- 01 / Subject
- a ceramic tea set with a brushed brass handle
- 02 / Action
- steam rising from the open pot
- 03 / Setting
- on a dark walnut counter in a quiet morning kitchen
- 04 / Frame
- waist-level three-quarter product view, generous space on the left
- 05 / Light
- soft window light from camera right, restrained shadow contrast
- 06 / Finish
- natural editorial photography, warm neutral color grade
Start with decisions
The anatomy of a useful AI image prompt
A generator cannot see the image in your head. It can only resolve the nouns, relationships, and visual constraints you place in the brief. Order the request from scene identity to surface treatment, so later details refine the picture instead of competing with it.
Begin an AI image prompt with the subject doing something. “A cyclist” leaves pose, direction, energy, clothing, and context open. “A rain-soaked bicycle courier checking a folded map under a shop awning” gives the model an event with visible evidence. The action also helps the body, clothing, and nearby objects agree with one another.
Next, place the event. Name a location only when its physical traits matter: a narrow tiled passage, a high-ceilinged studio, or a wind-cut coastal road. Describe foreground, subject plane, and background when placement matters. An AI image prompt that says “a red chair, a lamp, a window” has three nouns. One that says “the red chair faces the window, with the floor lamp just behind its left arm” has a composition.
Camera and light come after the scene is stable. Use framing that a photographer could set: eye level, low angle, top-down, close portrait, wide establishing frame, or three-quarter product view. Then name the light source and direction. “Soft light” is broad; “one large north-facing window from camera left, with gentle falloff across the background” is a decision.
Finish with medium, material, palette, and mood. These words should agree. Natural editorial photography, restrained warm neutrals, and lightly textured shadows reinforce one another. Mixing glossy 3D render, documentary film, watercolor, and hyperreal studio photography asks the model to negotiate four different images. Pick the treatment that carries the idea.
One brief, three levels
Add control only where the image needs it
Length is a consequence of complexity. The useful test is whether another person could sketch the same arrangement after reading the brief. These versions show how one idea gains control without filling the prompt with synonyms. A compact AI image prompt can still be precise when every phrase has a clear visual job.
Working brief / 39 words
Direct the frame
“Handmade ivory ceramic teapot with a brushed brass handle on a dark walnut counter, three-quarter product view at waist level, soft morning window light from the right, quiet editorial food photography, warm neutral grade.”
Controlled brief / 67 words
Protect the layout
“Create a natural editorial product photograph of a handmade ivory ceramic teapot with a brushed brass handle. Place it on a dark walnut counter, slightly right of center, with empty space on the left. Use a waist-level three-quarter view and one soft morning window from camera right. Keep the glaze matte, the reflections restrained, and the background kitchen softly out of focus. No text or extra tableware.”
Revision desk
Diagnose the output before adding more words
A weak result does not automatically need a longer AI image prompt. Compare the output with the brief and identify the first visible decision that broke. Repair that line, hold everything else steady, and generate again. This produces a useful test instead of a new lottery ticket. Mark the revised line in the AI image prompt so the next comparison has one known variable.
Revision rule
Change one class of variable per attempt: placement, light, surface, or style. If all four move at once, you cannot tell which edit improved the image.
The result feels generic
The brief names a category but gives no defining behavior or setting.
Edit: Replace “a beautiful cafe” with the exact counter material, time of day, seating pattern, and camera height.
Objects appear in the wrong place
The prompt lists objects without describing their relationships.
Edit: State “the cup sits in front of the pot; the spoon rests parallel to the saucer” instead of naming all three separately.
The style keeps drifting
Several visual media or conflicting eras compete for priority.
Edit: Choose one medium and one treatment. Keep “editorial product photography” and remove unrelated render or illustration labels.
A revision changes everything
The edit says what to change but never protects the accepted parts.
Edit: Name the locked subject, pose, framing, palette, and light before asking for one targeted change.
Model handoff
Keep the idea, change the instruction style
Treat the portable creative brief as source material. A Midjourney prompt can compress it into visual phrases and add an aspect-ratio parameter. A FLUX prompt benefits from complete sentences that spell out who is where. Stable Diffusion users may separate desired content from a short negative prompt, while ChatGPT Image and Nano Banana respond well to explicit creation or editing instructions.
Do not force platform syntax into the first draft. Write one readable AI image prompt, verify the visual logic, then translate that logic for the chosen model. Our reference analyzer performs that translation independently for General, Midjourney, FLUX, Stable Diffusion, ChatGPT Image, and Nano Banana, so you can compare how the same scene changes across six instruction styles.
Preflight
Read it once as a production brief
Check that the first sentence identifies the subject and action. Confirm that each secondary object has a position or role. Make sure the camera instruction can physically produce the requested composition, and name the source of light rather than relying on mood alone.
Read every adjective and ask what pixels it should change. “Premium” is an opinion. “Brushed aluminum, precise edge highlights, charcoal background, and controlled reflections” gives the model visible work.
Remove duplicates and conflicting media. Keep exclusions short and concrete. When editing an existing image, state the fixed identity, framing, text, and materials before the requested change. A careful AI image prompt makes acceptance criteria readable before generation begins.
If the source is a reference image, inspect it before writing. The AI Image Analyzer separates subject, environment, composition, lighting, color, camera feel, materials, style, and mood, which is useful when one vague impression needs to become a precise brief.
Save the accepted AI image prompt with the final output. The pair records which composition, light, and surface decisions produced the image you chose.
Questions from the prompt desk
AI image prompt FAQ
What should an AI image prompt include?
Start with the subject and action, then add the environment, composition, lighting, medium, mood, and any details that must remain fixed. Include only controls that change a visible decision. A short, specific brief is easier to revise than a long list of quality adjectives.
How long should an AI image prompt be?
There is no universal word count. A focused product scene may need 35 words, while a crowded editorial image may need 90. Stop when the subject, spatial relationships, camera position, light, and visual treatment are unambiguous. Extra synonyms often compete instead of helping.
Can the same prompt work in every image model?
The creative brief can travel, but its syntax should change. Midjourney benefits from concise visual direction and parameters, FLUX reads explicit natural-language relationships, Stable Diffusion workflows often separate positive and negative conditioning, and conversational editors need clear change and preservation instructions.
Why does the generated image ignore part of my prompt?
The request may contain competing styles, too many equally weighted objects, vague placement, or a camera instruction that conflicts with the composition. Move the defining subject and action earlier, state relationships plainly, and revise one variable per attempt.
Can I build a prompt from a reference image?
Yes. Use the Image to Prompt tool to analyze visible subject, environment, composition, light, color, camera feel, materials, style, and mood. The result is a new description inferred from the pixels, not recovery of the hidden original prompt, seed, or generation settings.
Begin with evidence
Turn a reference into six model-ready briefs
Upload one image to identify its visible choices and generate six prompt versions in one analysis. You can then edit the AI image prompt that fits your target model instead of starting with an empty text box.