2 prompts
Image Generation AI Prompts
Every image generation prompt in the library — tested, copy-ready, and tagged with the model it works best on. Free to use in personal or commercial work.
Image prompts fail differently from text prompts. A text model that misunderstands you produces something visibly wrong; an image model produces something confidently beautiful and not what you asked for, which is harder to diagnose and easier to accept.
The prompts here are structured around what image models actually respond to: subject, composition, lighting, medium and style, roughly in that order of influence.
How to use these
Order matters more than it does with text. Most image models weight the beginning of a prompt more heavily, so the subject goes first and stylistic detail follows. Adding "photorealistic, 8k, highly detailed" to the end of a poorly specified subject changes almost nothing — those words are doing far less work than their popularity suggests.
Change one variable at a time. The instinct when an image is wrong is to rewrite the whole prompt, which loses whatever was working. Adjust lighting, or composition, or medium — not all three.
What to watch for
Negative prompts and aspect ratio behave differently across models, so a prompt tuned on one will need adjusting on another. Text inside images remains unreliable everywhere, and counting is worse — asking for exactly five objects gets you approximately five.
On rights: what you may do commercially with a generated image depends on the model's terms and on your jurisdiction, and it is genuinely unsettled in places. Check the provider's licence before using output in anything you are paid for. Generating a recognisable person or a protected character is a separate question again, and one the model will not stop you from asking.