What is a negative prompt in AI generation?

A negative prompt lists what you don't want in the output — the counterpart to the prompt describing what you do.

They're most useful for recurring artefacts rather than subject matter: extra fingers, warped text, watermarks, blur, oversaturation. Listing "no cars" to keep cars out of a street scene works far less reliably than people expect, because models weight the presence of a concept more than its negation.

Support varies. Stable Diffusion and tools built on it expose negative prompts directly. Midjourney uses a `--no` parameter. Several newer models, including the ones behind Leaxor, don't expose them at all — the style is fixed by the pipeline rather than steered per generation, so there is nothing to negate.

If a tool you're using does support them, the highest-value entries are usually quality artefacts, not content exclusions.