How we test prompts

1. 01Write and review. A prompt starts as a draft. We check it against our content policy: no real people, nothing sexual or violent, and a clear own-photo notice when it edits a picture.
2. 02Run it on the named model. We paste the exact text you will copy into the model we list, in a fresh session, with default settings. If the prompt needs an uploaded photo, we use a photo we own.
3. 03Publish the real result. We export the image the model produced and publish it unedited: no retouching, no upscaling tricks, no stock images. We record the model, its version and the test date.
4. 04Note what made it work. We add short tester’s notes on the words that mattered and variations worth trying.
5. 05Re-test when models change. Models update often. When a result no longer matches, we re-test the prompt, replace the image and update the date — or remove the prompt.

What “tested” means on a prompt page

Each prompt page shows the model and version used, the test date, and the example image. Results from AI models vary from one run to the next, so your image will not be identical — but the same prompt should give you a result in the same spirit.

What we never do

• Publish a prompt we have not run.
• Show an image that was not produced by the prompt next to it.
• Inflate counts: the number of prompts in a page title is the number of tested prompts on that page.

Spotted a prompt that no longer works? Tell us and we will re-test it.