AI image generation prompts for Midjourney, Gemini & ChatGPT

How to turn a text prompt into the picture you had in mind

An image prompt is a description of a picture that does not exist yet, so the model fills every gap with its own guess. The prompts on this page close those gaps in a fixed order: the subject first, then the setting, the style or medium, the light, the camera angle or composition, and finally the shape of the frame. When you swap in your own subject, keep that order. A model reads the first words of a prompt as the most important, and a product shot that opens with the background instead of the product often comes back with the product too small.

The second thing to decide is whether the picture needs readable words. A poster, a thumbnail or a label with text is a different job from a wallpaper or a concept painting. Lettering is still the weakest part of most image models, which is why the text prompts here put the exact words in quotes and keep them short, and why many people add the headline afterwards in a design tool.

How to use these prompts

1. 1

   Pick the prompt closest to your goal and note the aspect ratio at the end. Change it first if you need a square post, a tall story or a wide banner.

2. 2

   Replace every [BRACKETED] part with something concrete: “a green ceramic mug with a chipped rim” works better than “a mug”.

3. 3

   Generate three or four versions before you change any words, so you can tell a bad prompt from a bad roll.

4. 4

   Change one thing at a time in the next round, such as the light or the camera angle, and keep the rest of the wording the same.

Get better results

• Name the medium. Words like “studio product photo”, “flat vector illustration”, “watercolour” or “3D clay render” change the result more than adjectives such as “beautiful” or “stunning”.
• Describe light like a photographer. Golden hour, soft window light, a single hard spotlight or overcast daylight each give a different mood, and the model follows them closely.
• Say what to leave out. In Midjourney the --no parameter removes things such as text or people. In Gemini and ChatGPT, write it as a plain sentence, for example “no people, no writing on the walls”.
• Leave room for your own text. For thumbnails, posters and ads, ask for empty space on one side and add the words yourself, which keeps spelling and fonts under your control.
• Keep a prompt log. Save the exact wording of every picture you like. Small changes in wording are hard to remember and easy to lose.

If the answer is not right

• The subject is too small. Move the subject to the start of the prompt and add a framing word such as “close-up” or “fills most of the frame”.
• Words are misspelled. Shorten the text to a few words in quotes, try a tool that handles lettering well, or remove the text and add it in an editor.
• Every result looks the same. Change the style word or the camera angle, not the subject. A new medium is the fastest way to a new look.
• Hands, faces or products look wrong. Regenerate a few times, simplify the pose or action, and crop or retouch small errors instead of starting again.

Each tool has its own rules on commercial use and on images of real people, brands and trademarks. Do not use a generated picture to stand in for a real product photo in a listing, and say that an image is AI-made where a platform asks for it.