Prompt Library

Every GPT Image 2.5 prompt here is printed in full beside the picture it actually produced, with the model, quality and size it ran at. Short on purpose — not nine thousand you have to sift through.

23 prompts · each one with the image it made

Nothing to upload

12

The prompt is the only input. Copy one, change a noun, run it in text to image.

Start from a picture you have

10

Each of these shows the before and the after, because an edit shown only as its output tells you nothing about what the edit did. Run one in image to image.

Three steps

How to use one of these

Taking a whole prompt and changing the subject works, and it is what most people do first. Taking one clause works better, and it transfers to subjects nothing like the original.

  1. Pick the job, not the picture

    Every card is named after a task, not a genre. Find the one whose job matches yours — the subject in the example does not matter, and swapping it is the easy part.

  2. Steal the constraint clauses

    The lines worth copying verbatim are the ones that forbid something: "no extra text", "do not change anything else", "no checkerboard". Those are what most people write worst and they transfer to any subject.

  3. Match the settings, or know why you did not

    Every card carries the model, quality and size the example ran at. A diagram at Draft is a diagram whose labels have gone soft — the settings are part of the prompt.

None of these is your idea? The prompt generator writes one from your own words — and hands you the settings to run it at.

The small print

Free to copy, free to run once

Every prompt is printed at full length, exactly as it was run — nothing truncated, nothing paraphrased, no watermark over the result. Copying costs nothing and needs no account. Only an image we actually render spends a credit.

Free here, no account

  • Reading and copying every prompt, in full
  • The settings each one was published at
  • Your first image — Standard, 1K, no card

Costs credits

  • HD and above, 2K and 4K
  • Sunburst, the quality model
  • Batches, and the 3:1 and 1:3 wide frames

Copy as many as you like, today, without telling us who you are.

Counted, not claimed

Where these came from

The three largest prompt collections on the web hold about nine thousand entries between them, and almost all of it is GPT Image 2 material published under GPT Image 2.5 headings — because none of them records which model made which picture. We record it, so here is the tally, counted from the library itself every time this page renders.

Published by OpenAI in the 2.5 prompting guide
23/23
Published by somebody else, credited on the page
0/23
Generated here
0/23
Made on a named GPT Image 2.5 model
23/23
Editing prompts shown with their input image
11/23

All 23 carry a source link and the model that produced them. Where a published example used a size we do not currently offer, the template says so rather than quietly resolving to a different frame.

Before you copy anything

Prompt library questions

Where these prompts came from, what the settings mean, and what running one costs

Are these GPT Image 2.5 prompts free to copy?

Yes. Every prompt is printed in full and free to take, edit and use. No account, no email, and copying never spends credits. Generating an image does.

Why are there only a couple of dozen?

Because choosing is the point. The large libraries publish thousands and hand you a filter, which leaves you doing the picking you came here to avoid. Every prompt here is one we would use.

Were these images actually made with GPT Image 2.5?

Every one says so on its own page, including which model. The first release is drawn from OpenAI's official GPT Image 2.5 prompting guide, where the prompt, the settings and the picture were published together and rendered on Flare and Sunburst. That is not the same as us having run it, so the cards say official rather than claiming it as ours.

Why does every prompt list a quality and a size?

Because on this model they change the result, not just the price. A diagram at Draft is a diagram with soft labels; the same prompt at HD works. The settings are part of the prompt, so they are printed with it.

Flare or Sunburst — does it matter which?

For most of these, no. Sunburst holds fine texture and dense small type slightly better, which matters on the portrait, the diagram and the slide. Each template names the one its example ran on. See Flare or Sunburst for the full comparison.

Why do the editing prompts show two pictures?

Because an edit shown only as its output is indistinguishable from a generation. If you cannot see the input, you cannot tell what the prompt did — which is why a library of edit prompts without before-and-after pairs is not much use.

Why does the rest of my image change when I only asked for one edit?

Because an edit generates a new image conditioned on the old one rather than patching a region. Everything is in play unless an instruction pins it, which is why every editing prompt here carries an explicit list of what must not change.

What does running one of these cost?

It is printed on every card, in credits, computed by the same code that debits your balance — including the surcharge an edit carries for its reference images. Your first image is free, at Standard and 1K, with no account.

Take one and run it.

Copying needs no account and nothing here is gated, truncated or watermarked. Your first image is on us — Standard, 1K, no card.

Written and maintained by Andy SwiftPublished Last updated

Your first GPT Image 2.5 image is free

Generate my first image