AI Image Prompting Guide: Examples, Editing, and Model Tips

A useful image prompt describes a picture someone could actually draw: what is in it, how it is framed, what the light looks like, and what matters most. Start with that brief, choose an image-capable model, then inspect the result. Adding a long list of impressive adjectives is not a substitute for deciding what you want.

Want a starting point? Browse illustrated prompts for posters, product photos, character sheets, and social visuals. The library is public; generating images requires sign-in and a Pro plan.

Browse image prompt examples

Quick Takeaways

Name the subject, medium, composition, lighting, and important constraints.
Use the model's supported settings for dimensions and output quality.
For edits, explain both the change and the details to preserve.
Check lettering, anatomy, perspective, and unwanted objects before using an image.

A practical AI image prompt formula

Use subject + medium + composition + lighting + constraints. Add only details that would change your decision about whether the result is usable. A logo mockup needs accurate placement and clear lettering; a landscape illustration may need a particular palette and mood.

Reusable brief

Create [medium] of [subject] in [setting]. Use [framing and placement], [lighting], and [palette]. Preserve [important details]. Leave [space needed for your design]. The image should contain [specific contents].

Compare "a nice coffee ad, stunning, best quality" with "a close product photo of a cream mug on a wooden counter, morning light from the left, the mug on the right, clear space on the left for a headline." The second brief gives you concrete things to check. It does not guarantee the model will follow every instruction.

Begin with a small set of requirements. If the first result misses the crop, fix the crop instruction before adding more props, materials, and effects. Keep a record of the model and settings with a prompt you plan to reuse; the same words can behave differently in another tool.

AI image prompt examples you can adapt

These are example briefs, not benchmark results or promises of a specific output. Replace the subject, colors, and layout with your own requirements. Select the aspect ratio in the tool when that setting is available.

Product photography

Create a product photo of a matte cream ceramic mug on a pale oak table. Three-quarter view, soft window light from the left, neutral beige background, realistic ceramic texture. Put the mug on the right half and leave the left half clear for a headline. The mug is unbranded and the scene contains only the mug and table.

Review: Check the handle, rim, shadow direction, and whether the headline area is actually clear.

Social media illustration

Create a flat editorial illustration of a person organizing a weekly calendar at a desk. Limited palette of navy, cream, and warm orange. Bold simple shapes, clean outlines, generous empty space above the subject. Show calendar blocks as plain colored rectangles without writing.

Review: Check that the calendar stays readable at thumbnail size and that no accidental text appears.

Poster with short text

Design a portrait poster for a fictional weekend pottery workshop. Off-white background, one terracotta bowl centered in the lower half, soft shadows. At the top, place the exact words "Make something" on one line and "Saturday pottery" below in smaller type. Keep all lettering inside wide margins. Include no other words.

Review: Read every letter and inspect the margins. Add final typography in an editor if spelling or spacing remains wrong.

Landscape scene

A watercolor illustration of a quiet mountain lake at sunrise. Pine branches frame the foreground; a small wooden cabin sits on the far shore. Muted blue and warm peach palette, light mist above the water, visible paper grain, wide composition.

Review: Check the cabin perspective and the reflection instead of accepting the image just because the colors look good.

Edit an existing image

Using the attached product photo, change only the backdrop to a pale warm-gray studio background. Preserve the product shape, color, label wording, camera angle, and surface texture. Keep the original crop and add a soft contact shadow beneath the product.

Review: Use this with an image-editing model and your own reference photo. Compare the label and product silhouette with the original.

How prompting changes between image models

GPT Image: describe the job and review the details

Use a direct brief with the subject, layout, and details that must be preserved. For edits, provide a reference image and specify what should change. Output size and quality are tool settings, not guaranteed by adjectives in a prompt. OpenAI documents remaining limitations with text, recurring characters, and precise composition, so inspect those closely.

Gemini / Nano Banana: make the scene and edit clear

Describe the scene in connected sentences. For an edit, explain the intended change and what must stay the same. Google recommends specific context, camera language, and positive descriptions of the desired scene. Model and interface capabilities vary; select an image-capable model before trying an image workflow.

Midjourney: keep the visual description concise

Midjourney recommends short descriptions of the subject, medium, light, and composition. Its parameters belong at the end of the prompt. For example, --ar 16:9 requests a wide aspect ratio in Midjourney. Use its supported settings rather than copying old version numbers from a tutorial. Midjourney is discussed here for comparison; it is not listed in Chocolatey AI's image model selector.

Stable Diffusion: check your model and interface

Some Stable Diffusion interfaces provide a separate negative-prompt field. Weighting syntax and available controls depend on the interface; do not assume parentheses or weights work in every image model. A useful starting point is a clear subject, medium, and composition, followed by only the exclusions you actually need. Stable Diffusion is also covered as a comparison, not as an available Chocolatey AI model.

Portability rule: keep the visual brief when switching models, but remove tool-specific flags. A Midjourney parameter is not a GPT Image or Gemini instruction.

How to fix an AI image prompt that is not working

The subject looks generic: Replace broad adjectives with visible properties: material, color, shape, pose, and a specific setting. Decide which two details matter most.
The layout leaves no room for your headline: Specify which side contains the subject and which side stays empty. Ask for a simple backdrop. If placement still fails, use an editor to finish the layout.
Text is misspelled: Shorten the wording, quote the exact text, and specify its location. Inspect every letter. For precise brand typography, place the final text yourself.
An edit changes the product or character: Use the original image as a reference and list the details to preserve. Request one change at a time, then compare the result against the original.
You get unwanted objects: Describe the desired contents positively, such as "an empty tabletop with one mug." Use a separate negative-prompt control only when your chosen interface supports it.
The image is too small for the job: Check the actual pixel dimensions and the available output settings. Writing "8K" in the description does not set an 8K export. Prepare a higher-resolution output through supported settings or another suitable workflow.

Does JSON make image prompts better?

JSON can help you organize and reuse a brief. It is not a universal image-generation protocol, and valid JSON does not guarantee accurate lettering, consistent characters, or precise placement. Start with plain language. Use structured fields if they make your own workflow easier to maintain, and verify what your interface accepts.

Review the image before you use it

Compare the result against your original brief, not just its overall appearance.
Read every label and check hands, faces, repeated objects, reflections, and perspective.
Confirm the crop, margins, pixel dimensions, and file format for the intended destination.
Use reference images you have permission to use and follow your platform's publishing rules.
Keep human review for product labels and other details that must remain accurate.

Save a successful prompt together with its output and settings. That gives you an example to refine next time. A good prompt library should help you decide what to change, rather than encourage copying words you do not understand.

Try a prompt in Chocolatey AI

Open the public image prompt library, choose an example, and adapt its subject, palette, and composition. Use the image model selector to see available models. GPT Image and Google image models are listed; the Midjourney and Stable Diffusion guidance above is for comparison.

For broader instruction-writing techniques, read our guide to prompting AI effectively. If a long conversation starts losing important details, see how model context windows work.

Editorial review: October 11, 2026. Model guidance was checked against OpenAI's image-generation documentation, Google's Gemini image guide, Midjourney's Prompt Basics, and Poe's image-prompt guidance. The examples are original illustrative briefs, not measured comparisons.

Originally published November 8, 2025 by the Chocolatey AI Team.