AI image prompts are not interchangeable between tools. Gemini’s guidance favors a full descriptive paragraph, Midjourney’s favors a short phrase plus parameters, and ChatGPT takes plain-language instructions you can refine in the chat. This guide takes one picture idea and writes it for each of the three, so you can see what to keep and what to change when you switch tools.
Each version below is an original Promptline example built from the vendors’ own prompting advice, linked next to each claim. They are starting points to adapt, not tested results.
On this page
Start from one brief, then translate it
Before you write for a particular tool, settle the decisions that belong to the picture itself. These carry over everywhere:
| Decision | Example for this guide |
|---|---|
| Subject | One ceramic teapot with a chipped spout |
| Setting | A wooden kitchen shelf by a window |
| Medium | Colored-pencil illustration |
| Light and color | Morning light from the left; muted greens and cream |
| Composition | Teapot in the right third, empty wall on the left |
| Purpose | A blog header that needs space for a headline |
Midjourney’s Prompt Basics lists much the same areas to consider: subject, medium, environment, lighting, color, mood and composition. What changes between tools is how you phrase those decisions and which settings live outside the sentence.
Gemini: describe the scene in full sentences
Google’s prompting guide for its image model starts with one principle: “Describe the scene, don’t just list keywords.” It says a narrative, descriptive paragraph will almost always produce a more coherent image than a list of disconnected words (Google Developers Blog). The same post recommends stating the purpose of the image and phrasing exclusions positively, for example “an empty, deserted street” instead of “no cars.”
A colored-pencil illustration of a single ceramic teapot with a
small chip on its spout, standing on a worn wooden kitchen shelf
beside a window. Soft morning light comes from the left and
catches the glaze. The palette is muted sage green and warm cream.
The teapot sits in the right third of the frame; the left side is
a plain, empty wall, because the image will be a blog header with
a headline placed there later. Wide landscape format.
The purpose sentence explains why the left side stays empty. Gemini’s image model is also built for conversational editing, so a follow-up such as “keep everything the same, but make the light warmer” is the expected way to refine. For more Gemini-specific templates, see the Nano Banana prompts guide.
Midjourney: a short phrase, then parameters
Midjourney’s documentation takes the opposite line on length: “Short and simple prompts typically generate the best images,” and long lists or detailed instructions “can confuse the process” (Prompt Basics). Its own example turns a chatty request into “Colored pencil illustration of bright orange California poppies.”
Colored pencil illustration of a chipped ceramic teapot on a
wooden kitchen shelf, morning window light, muted sage and cream,
empty wall on the left --ar 16:9
Three habits from the same page apply here:
- Use specific numbers or collective nouns: “three teacups” rather than “teacups.”
- Describe what you want instead of what you don’t. Midjourney notes that “no cake” may still produce a cake, and points to its
--noparameter for exclusions. - Put settings such as aspect ratio at the end as parameters, not in the sentence.
The purpose (“a blog header”) is left out because Midjourney reads the prompt as a picture description, not as a conversation. Our Midjourney parameters cheat sheet covers --ar, --stylize and the other settings.
ChatGPT: plain instructions you can edit afterwards
OpenAI’s help article Images in ChatGPT says you can simply describe the image you want ChatGPT to create, and that ChatGPT Images can follow instructions to add text, add details or make the background transparent. You can also upload an image and describe changes, or use the selection tool to edit one area.
Create a wide colored-pencil illustration for a blog header.
Show one ceramic teapot with a small chip on its spout on a
wooden kitchen shelf next to a window, lit by soft morning light
from the left. Use muted sage green and cream. Place the teapot
in the right third and keep the left side as a plain wall, with
no text, so I can add a headline later.
Instructions read naturally here, including the reason for the empty space and a direct “no text” request. If one area comes out wrong, select it and describe only that change rather than regenerating the whole picture. The ChatGPT image prompts guide collects more of OpenAI’s examples.
What to change when you move a prompt between tools
| If your prompt came from | Before using it in Gemini | Before using it in Midjourney | Before using it in ChatGPT |
|---|---|---|---|
| Gemini (long paragraph) | — | Cut to the key phrase; move format to --ar | Keep the sentences; add the purpose if missing |
| Midjourney (short phrase) | Expand into full sentences; add purpose and light | — | Rewrite as instructions; remove -- parameters |
| ChatGPT (instructions) | Turn requests into scene description | Strip “create”, “I want” and reasons | — |
Parameters such as --ar or --no mean nothing to Gemini or ChatGPT, and a long paragraph pasted into Midjourney works against its own advice. Translating the wording is usually faster than fixing a result afterwards.
Check the result against the brief, not the prompt
Compare the image with the decisions in your brief: is there one teapot, is the left side clear, is the light coming from the left? Change one thing per follow-up so you can tell which instruction caused which result. A general brief-writing method for illustrations and edits is in our image prompts guide, and camera-style briefs are in the photo prompts guide.
Questions
What makes a good AI image prompt?
A clear subject, medium, setting, light, composition and purpose. Gemini and ChatGPT take these as sentences; Midjourney takes them as a short phrase plus parameters.
Can I use the same image prompt in ChatGPT, Gemini and Midjourney?
You can reuse the decisions, but the wording should change. Midjourney’s docs recommend short prompts, while Google recommends a descriptive paragraph.
How do I stop an AI image tool from adding something I don’t want?
Describe the scene positively (“a plain empty wall”) as Google suggests. In Midjourney, use the --no parameter instead of writing “no” in the prompt.
Where do aspect ratios go?
In Midjourney, as --ar at the end of the prompt. In Gemini and ChatGPT, say the format in words or use the interface’s own setting where one exists.
