What is a Negative Prompt?
Positive prompts describe what you want; negative prompts describe what you don't. Both run through the same text encoder and guide the diffusion process in opposite directions. This gives much finer control than positive prompts alone — it's often easier to fix a problem by saying 'no blurry, no deformed hands' than to craft a positive prompt that avoids those outcomes.
The most useful negative prompts tackle common AI failure modes: deformed faces and hands, low resolution, watermarks, cropping artifacts. Power users maintain a 'quality negative prompt' they reuse on every generation.
How it works
Dual guidance
During each denoising step, the model generates predictions for both positive and negative prompts. The final step extrapolates away from the negative prediction and toward the positive one.
Common quality boosters
A standard 'quality' negative prompt: 'blurry, low quality, low resolution, watermark, signature, deformed, ugly, extra fingers, missing fingers, extra limbs, bad anatomy'.
Common use cases
- Removing common artifacts (blur, watermarks, low resolution)
- Fixing anatomical errors (extra fingers, deformed faces)
- Avoiding specific concepts (e.g., 'no cartoon style' in a photorealism workflow)
- Boosting perceived quality with a reusable 'quality negative' template
