It is a second prompt running in the opposite direction. The model computes what the positive text pulls towards and what the negative text pulls towards, and pushes away from the second. That is why it works where a negation in the main prompt does not: "no text" in a positive prompt still puts the concept of text into the mix, and models frequently render it anyway.
It is most useful against recurring, describable failures — watermarks and stray captions, extra limbs, a plastic sheen you keep getting. It is close to useless as a quality incantation. A stack of vague terms like "bad, ugly, low quality" mostly narrows the space the model can work in without telling it anything specific.
Not every model exposes one. Where the field is absent, the model was either trained without that mechanism or the provider chose not to surface it, and the equivalent move is to describe the thing you do want more precisely.
In practice
- Name concrete nouns you keep seeing, not moods you dislike.
- Keep it short. A negative prompt longer than the prompt is usually fighting the prompt.
- If a term appears in both prompts, the two cancel and you have wasted the slot.
The mistake to avoid
Copying a hundred-word negative prompt from a forum. Most of those terms do nothing on your model and a few actively suppress detail you wanted.
Where you will run into it
- Text to Image Generator — One prompt. Every image model. One studio.
Related terms
Prompt
A prompt is the written instruction a generative model reads to decide what to make — the one input almost every model requires.
Prompt adherence
Prompt adherence is how faithfully a model does what the prompt actually said, as opposed to producing something attractive in the same neighbourhood.
CFG scale
CFG scale controls how strictly a model obeys your prompt, trading obedience against the model's own sense of what a natural image looks like.
Safety checker
A safety checker is an automated filter that inspects prompts, outputs or both and blocks material a provider does not permit.
Seed
A seed is the number that decides the random starting noise for a generation, so the same seed with the same settings reproduces the same output.
The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync in a single app.