Negative Prompts: What to Actually Put in Them

# Negative Prompts: What to Actually Put in Them If you use Stable Diffusion or Flux, you have a negative prompt field — a place to list what you *don't* want. Most people either ignore it or stuff it with a hundred random words copied from a forum. Both are mistakes. Here's how to use it well. ## What a negative prompt actually does It pushes the model *away* from concepts. "blurry" in the negative prompt makes the model less likely to produce blur. It's a steering tool, not a magic quality button — and like the positive prompt, precision beats volume. ## What belongs in it **Technical defects you keep seeing.** The classic, genuinely useful entries: `blurry, low resolution, jpeg artifacts, watermark, signature, text`. These target real failure modes. **Anatomy problems (for people).** `extra fingers, deformed hands, extra limbs, fused fingers, mutated`. Hands are still where models struggle most; these help. **Things specific to your image.** This is the part people miss. If your generations keep adding a hat you don't want, put `hat` in the negative. If backgrounds keep coming out cluttered, add `busy background, clutter`. The best negative prompt is *reactive* — you add to it based on what's actually going wrong in your results. ## What does NOT belong in it **A giant copy-pasted wall of words.** Those 80-term negative prompts floating around forums mostly add noise. Many terms contradict each other or target things that were never going to appear. A focused 8–12 term negative prompt outperforms a bloated one. **Style you could just not ask for.** Don't put "cartoon" in the negative if you simply want realism — ask for "photorealistic" in the *positive* prompt. Negatives are for suppressing defects and unwanted intrusions, not for setting your main style. ## A solid starter negative prompt For realistic images: > blurry, low resolution, jpeg artifacts, watermark, text, signature, deformed hands, extra fingers, extra limbs, oversaturated Then **add to it reactively.** Generate, look at what's going wrong, and put *that* specific problem in the negative. Over a few iterations you'll build a negative prompt tuned to your exact style and subject — which is worth far more than any copied list. ## Models without a negative field DALL-E 3 and Midjourney handle this differently — Midjourney uses `--no X`, and DALL-E wants you to phrase things positively ("a clear, uncluttered background") since it has no negative field at all. If you're moving between models, an [image-to-prompt tool](/tools/image-to-prompt) is handy for seeing how the same intent gets expressed in each one's syntax. The rule of thumb: keep it short, keep it specific, and grow it in response to what you actually see — not what a forum told you to paste.