A negative prompt is an instruction that tells an AI generation model what to exclude from the output, specifying unwanted elements, qualities, or characteristics rather than describing what should be present. While a standard prompt describes the desired result, a negative prompt works alongside it to steer the model away from common failure modes, visual artifacts, or content that would be inappropriate for the intended use.

In image generation systems that support negative prompting, common entries include terms like "blurry," "low quality," "distorted," "watermark," "text," "extra limbs," or "bad anatomy," which guide the model away from known generation weaknesses. More contextually specific negative prompts might exclude particular styles, visual elements, or moods that conflict with the creative intent - a creator aiming for a clean, minimal aesthetic might use negative prompts to exclude clutter, complex backgrounds, or busy textures. The technical mechanism varies by model and implementation, but the general principle is that the model steers its generation away from concepts associated with the negative prompt terms while moving toward those in the positive prompt.

Not all AI generation tools expose negative prompting directly to users, and the effectiveness of negative prompts varies significantly between models and implementations. When available, negative prompting is most useful for addressing recurring problems in a particular model's outputs or for quickly communicating what a generation should not look like when describing what it should look like would be difficult alone.