ChatGPT Image Prompts That Work: A Copy-Ready Library

ChatGPT is very good at generating images, but only if you tell it what you actually want. A one-line prompt gets you a generic stock-looking picture. A prompt that names the subject, style, framing and format gets you something you can publish.

This page is a library of prompts I keep coming back to, grouped by what you need the image for. Each one has a copy button and a short note on why it works. Change the parts in brackets and paste.

How a good image prompt is built

Most good image prompts answer the same seven questions. You don't need all of them every time, but every one you skip is a decision the model makes for you.

Two habits make the biggest difference. First, put any text you want inside quotes, and keep it short; long sentences come out misspelled. Second, say what you don't want: "no text", "no watermark", "no extra people".

Blog and social media

Create a wide 16:9 cover image for a blog post about [topic]. Flat vector illustration of [main object, e.g. a laptop with a terminal window], placed on the left third. Keep the right half mostly empty with a plain [color] background so a title can be added later. Palette: [2–3 colors]. Clean, modern, not cartoonish. No text, no logos, no watermark.

Leaving empty space on one side gives you room to add the title yourself, which is more reliable than asking for text.

Create a 16:9 thumbnail. On the left, [subject with a clear emotion, e.g. a surprised developer looking at a laptop]. On the right, the text "[3 words max]" in a very bold sans-serif, white with a thick dark outline, large enough to read on a phone. High contrast, simple background in [color]. No other text.

Thumbnails are viewed tiny, so the prompt forces three words at most and one strong subject.

Create a vertical 9:16 image of [subject]. Keep the important content in the middle 60% of the frame, leaving the top and bottom fifth calm and uncluttered. Style: [photo / illustration]. Bright, high contrast, readable at small size. No text.

Vertical apps cover the top and bottom of the screen with buttons, so the subject is kept in the middle.

Product, portfolio and app images

A realistic studio product photo of [product], centered on a [material, e.g. light oak] surface. Soft diffused light from the top left, gentle shadow on the right. Plain [color] background. Shot at eye level with a slight angle, shallow depth of field. No text, no branding.

Naming the light, the surface and the camera angle is what separates a product photo from a random render.

A modern smartphone with no visible brand, held in one hand, slightly angled toward the viewer. The screen shows a clean [type of app, e.g. task manager] interface in [colors]. Blurred home-office background, natural daylight. Photorealistic. No logos.

Asking for a generic phone avoids real brand logos, and you can paste your real screenshot on top later.

Illustrations and icons

Create a set of [number] icons in one image, arranged in a grid on a white background: [list, e.g. database, server, cloud, lock, terminal, git branch]. Style: simple line icons, 2px rounded stroke, [color], same size and padding for every icon. No text labels.

Describing the grid, stroke and color once keeps all icons consistent, which is the hardest part of an icon set.

An isometric illustration of [concept, e.g. three servers connected to a database and a laptop], on a plain [color] background. Soft pastel palette with [accent color] highlights, clean geometric shapes, subtle shadows. Leave some empty space around the edges. No text.

Isometric views make technical ideas look tidy and work well as article headers.

Fixing a result without starting over

When an image is almost right, don't rewrite the whole prompt. Ask for one change and say what must stay the same.

Keep this image exactly the same, including the composition, colors and style. Only change [one thing, e.g. the laptop screen to show a code editor instead of a terminal].

Naming what stays the same stops the model from regenerating everything.

If it keeps changing things you didn't ask for, that's usually the point to move the image to an editor. In my case that's Gemini; the Gemini image editing prompts guide has the prompts I use there.

Getting more out of the free plan

The free plan only gives you a few images, so each one should count:

• Write the full prompt the first time. Fill in all seven parts above instead of starting vague and correcting it.
• Ask for one change per message when you iterate, so you don't burn a generation on a result that changed three things.
• Generate the base, edit elsewhere. Use ChatGPT for the hard part (the composition and style) and do small edits in another tool.
• Save prompts that worked. A prompt that gave you a good result once will usually work again for a similar image.

FAQ

Why does ChatGPT get the text in my image wrong?

Long or unusual text is still the weakest part of image models. Keep the text to a few words, put it in quotes, and if it still comes out wrong, generate the image without text and add the words yourself in any editor.

Can I use images made with ChatGPT commercially?

OpenAI's terms let you use the images you create, including commercially, as long as you follow their usage policies. Avoid real people's likenesses, logos and trademarks, and check the current terms before using an image in a product.

Is ChatGPT or Gemini better for images?

In my experience, ChatGPT is better at creating a new image from a detailed description, and Gemini is handy for quick edits to an image you already have. Many people use both, which is the workflow on this page.

What aspect ratio should I ask for?

Ask for the format you'll publish in: 16:9 for blog covers and YouTube, 9:16 for TikTok, Shorts and Reels, 1:1 or 4:5 for Instagram posts.

Further reading

• Next: Gemini image editing prompts
• Related: How to write better prompts for AI coding assistants
• Related: Learning to code with AI