These 14 AI image prompts work with the main image generators, GPT Image (in ChatGPT and the OpenAI API), Google's Gemini image models (Nano Banana) and Midjourney, and each one follows that vendor's own prompting guide. They cover photorealistic scenes, product shots, illustrations, graphics with exact text, edits and Midjourney parameters, plus one prompt that asks ChatGPT or Claude to write image prompts for you.
How to use these prompts
Replace the blanks in double braces, or keep the default after the colon: {{subject:a ceramic coffee mug}} means that subject unless you change it.
Most prompts work in any of the three tools. Where a prompt depends on one tool's features (Midjourney parameters, transparent backgrounds in the OpenAI API), the heading says so.
Every prompt here describes original subjects. Don't fill the blanks with real people's names or copyrighted characters; see the last section for why.
Which models are current changes often. As of October 2026, OpenAI's image prompting guide covers GPT Image 2.5 and GPT Image 2 (with reference notes for older models), and says gpt-image-1.5 is scheduled to shut down on December 1, 2026. Google's Gemini image generation docs list the Nano Banana models and describe Imagen as "Google's legacy text-to-image models (shut down)." Midjourney's Version page lists V8.2 as the default version since July 24, 2026.
What do the official guides say makes a good image prompt?
The three vendors agree on more than they differ. Here is the core advice from each:
| Advice | GPT Image (OpenAI) | Gemini (Google) | Midjourney |
|---|---|---|---|
| Structure | "organize the prompt as scene, subject, details, and constraints" for complex requests | Describe the scene in rich detail; break complex scenes into steps | Short, simple prompts; a checklist of subject, medium, environment, lighting, color, mood and composition |
| Leaving things out | State constraints explicitly | Describe the scene positively ("semantic negative prompts") | Describe what you want; --no as a negative weight |
| Text in images | "Put required wording in quotes and describe its position and typography." | Be clear about the text, the font style and the overall design | Put the words in double quotation marks and keep them short |
| Photo realism | Ask for a photorealistic image or real photograph explicitly; camera specs are cues, not guarantees | Use camera and lens language: wide-angle shot, macro shot, low-angle perspective | Name the medium and lighting |
| Editing | "change only X" and list what must stay the same | Small conversational follow-ups | Edit model and image prompts |
Google's guide sums up the shared principle: "Be hyper-specific: The more detail you provide, the more control you have." It also recommends you "Provide context and intent: Explain the purpose of the image."
Which prompts make photos and product shots?
1. Photorealistic scene
Built on the photorealistic template in Google's guide, and works as well in GPT Image.
A photorealistic {{shot:wide-angle shot}} of {{subject:a small neighborhood bakery at opening time}} in {{setting:a quiet city street after rain}}.
{{light:Soft early morning light from the left, warm light glowing from the shop window}}.
Shot from {{angle:eye level}} with a {{lens:35mm lens}}, shallow depth of field.
Purpose: {{purpose:a hero image for a local food blog}}. Aspect ratio {{ratio:16:9}}.
No people's faces in focus, no visible brand logos.
2. Studio product photo
For clean ecommerce and catalog images of your own product. Google's guide offers a "high-resolution, studio-lit product photograph" template for exactly this.
A high-resolution, studio-lit product photograph of {{product:a matte black stainless steel water bottle with a bamboo lid}} on {{surface:a seamless light gray background}}.
Lighting: {{lighting:softbox from above and slightly left, subtle reflection on the surface}}.
Camera: {{camera:straight-on, slightly above product height}}, product centered, filling about 70 percent of the frame.
Show the product exactly as described. No added text, logos, props or extra objects.
3. Lifestyle shot from your own product photo
Upload a photo of your product as a reference image. Both GPT Image and Gemini accept image input for edits and composites.
Use the attached photo of our product as the exact reference. Keep its
shape, color, materials, proportions and any printed details unchanged.
Place it in this scene: {{scene:on a wooden desk next to a laptop and a notebook, morning light from a window}}.
Style: {{style:natural lifestyle photography, realistic, muted colors}}.
The product is the clear focal point. No other branded items, no text.
Aspect ratio {{ratio:4:5}}.
Check the result against the real product before using it to sell. A generated image that changes the product's look can mislead buyers.
Which prompts make illustrations and graphics?
4. Blog or article header illustration
Google's guide says to describe the artistic style, subject and medium, and to be specific about visual details for consistent results.
A {{style:flat vector illustration}} for an article titled "{{title:How small teams plan a product launch}}".
Show {{concept:three people around a large wall calendar moving sticky notes}}, drawn as simple original characters.
Visual details: {{details:bold outlines, limited palette of orange, cream and charcoal, generous white space}}.
Wide format, {{ratio:16:9}}, with empty space on the right for a headline. No text in the image.
5. A consistent icon set
Ask for the whole set in one image so the style matches, then crop.
A set of {{count:6}} icons on one canvas, arranged in a {{grid:3 by 2}} grid with even spacing.
Icons: {{icons:calendar, chat bubble, folder, magnifying glass, lightbulb, checkmark}}.
Style: {{style:2px rounded line icons, single color charcoal on white, no fills}}.
Every icon uses the same line weight, corner radius and size. No text labels, no shadows.
6. Poster or social graphic with exact text
Text is where image models make most mistakes, so OpenAI's guide says to quote the wording, describe placement and typography, ask for no extra text, and check spelling.
Create a {{format:square social media graphic}} for {{purpose:a free community workshop}}.
Text, exactly as written, nothing else:
- Headline at the top, large bold sans serif: "{{headline:Intro to Bookkeeping}}"
- Below it, smaller: "{{details:Saturday 14 November, 10am, Central Library}}"
Background: {{background:a simple illustration of an open ledger and a coffee cup, warm paper texture}}.
Leave clear margins around the text. Do not add any other words, logos or watermarks.
Always proofread every word in the output. If a date or name is wrong, regenerate or add the text in a design tool instead.
7. Background with space for text
Google's guide describes minimalist, negative space designs as useful "for websites, presentations, or marketing materials where text will be overlaid."
A minimalist composition featuring a single {{subject:sprig of eucalyptus}} positioned in the {{position:bottom right}} of the frame.
Background: {{background:soft off-white with a subtle paper texture}}.
At least two thirds of the image is empty space for text. Soft, even lighting, gentle shadow.
Aspect ratio {{ratio:16:9}}. No text.
8. Diagram or simple infographic
Good for slides and docs. Keep the labels short and check them.
A clean, flat infographic explaining {{topic:how a package moves from warehouse to doorstep}} in {{steps:4}} steps, left to right.
Steps and labels, exactly: {{labels:"Picked", "Packed", "Shipped", "Delivered"}}
Each step has a simple icon above its label, connected by arrows.
Style: {{style:white background, navy and orange accents, rounded sans serif labels}}.
No extra text, numbers or decoration beyond what is listed.
Which prompts edit an image?
9. Change one thing, keep the rest
OpenAI's guide recommends the "change only X" pattern and listing what must stay the same, because details are preserved by stating them, not by assuming.
Edit the attached image. Change only this: {{change:make the sky a clear blue late-afternoon sky}}.
Keep exactly the same: {{keep:the building, the people and their positions, the camera angle, the colors of the shop sign, the overall lighting direction}}.
Match the original's style, lighting and perspective. Do not add or
remove any objects.
10. Refine a result step by step
Both OpenAI and Google recommend small follow-ups instead of rewriting the whole prompt. OpenAI's guide: "Pass the previous output as the next edit input, request one change, and repeat the details to preserve."
That's close. Keep everything the same, but {{change:make the lighting a little warmer and move the mug slightly left}}.
Still keep: {{keep:the background, the mug's color and shape, the camera angle, the aspect ratio}}.
11. Transparent background asset (OpenAI API)
For logos, stickers and cutouts. OpenAI's guide says to request a transparent background with PNG or WebP output and to check the alpha channel, since "A drawn checkerboard is not transparency."
An original sticker illustration of {{subject:a cheerful cartoon cactus wearing sunglasses}}.
Style: {{style:thick white outline, bold flat colors, slight drop shadow}}.
Centered, with nothing else in the image. Transparent background.
Original design only: no existing characters, logos or trademarks.
When calling the API, set the background parameter to transparent and the output format to PNG or WebP; the words in the prompt alone may not be enough.
Which prompts are written for Midjourney?
12. Short Midjourney prompt with parameters
Midjourney's Prompt Basics favors short, clear prompts, and its parameter docs say parameters go at the end. Aspect ratio uses --ar with whole numbers, and the Stylize value defaults to 100 on a 0 to 1000 scale.
{{subject:three paper lanterns floating over a calm lake}}, {{medium:soft watercolor}}, {{lighting:dusk glow}}, {{mood:quiet and peaceful}} --ar {{ratio:3:2}} --stylize {{stylize:100}}
Raise stylize for a more artistic look, lower it to follow your words more literally. Change one thing at a time so you can see what each change did.
13. Keep things out of a Midjourney image
Midjourney's No parameter page says --no works like a negative weight of -0.5, and that its moderation reads each word after --no separately. Describe what you want first, and use --no for single, simple words.
{{subject:an empty mountain road at sunrise}}, {{details:fresh snow on the pines, long shadows across the asphalt}}, {{medium:cinematic photograph}} --ar {{ratio:16:9}} --no {{exclude:cars, text}}
Which prompt writes image prompts for you?
14. Turn a rough idea into prompts for each tool
Paste this into ChatGPT or Claude when you know what you want but not how to phrase it.
I need an image for: {{purpose:the header of a newsletter about home gardening}}
Rough idea: {{idea:hands planting seedlings, hopeful, spring}}
Tools I use: {{tools:GPT Image, Gemini and Midjourney}}
Write one prompt per tool, following each vendor's guidance:
- GPT Image: labeled sections for scene, subject, details and
constraints; quote any exact text.
- Gemini: one descriptive paragraph, positive phrasing instead of "no
X", camera or style language, and the purpose of the image.
- Midjourney: a short phrase covering subject, medium, lighting and
mood, then parameters at the end (--ar and, only if needed, --no).
Use only original subjects: no real people's names or likenesses, no
copyrighted characters, no brand logos. Then tell me which details to
change first if the result misses.
What should you not ask an image model to make?
Leave out real people's likenesses and copyrighted characters. Midjourney's Community Guidelines bar imagery of real people that could be used to harass, abuse, defame or otherwise harm, and OpenAI's image prompting guide tells you to "Keep designs original and non-infringing." Beyond platform rules, using someone's face or a protected character in marketing can create legal problems, so describe an original character or a generic person instead.
Don't present generated images as real evidence. A product photo, before-and-after picture or event photo that shows something that didn't happen can mislead customers. Google's Gemini image docs also note that "All generated images include a SynthID watermark."
Check every detail you plan to rely on: text, product features, hands, logos and anything a customer might take as a promise.
Keep your image prompts in one place
Once a prompt gives you the look you want, the exact wording matters, and it's easy to lose in a chat history. Kipt keeps prompts as Markdown in a private GitHub repo you own, with their blanks ready for the next subject; tag them by tool (GPT Image, Gemini, Midjourney) or by use (product, blog, social) and press ⌘K to find one. To give Claude or Codex your house image style as a reusable skill, paste prompt 14 with your style rules into the free SKILL.md generator. For text prompts in Google's assistant, see the Gemini prompts.