Mastering Negative Prompts:
What NOT to Say to Your AI
Negative prompts are the hidden superpower of AI image generation. Most users ignore them entirely — and it shows in the blurry hands, extra fingers, and muddy compositions they get back. This guide fixes that.
Every time you generate an AI image, you're having a conversation with a model that has seen billions of images and has strong statistical tendencies. Without guidance, it will drift toward the average — the median aesthetic of everything it has learned. Negative prompts are how you steer away from that median and toward something intentional and precise.
Used correctly, negative prompts can eliminate the most common AI image artifacts: malformed hands, extra limbs, watermarks, low-resolution textures, unnatural proportions, and the dreaded "AI face" — that uncanny, overly smooth look that immediately signals "this was generated by a machine."
What Are Negative Prompts?
In most AI image tools, you have two prompt fields: a positive prompt (what you want) and a negative prompt (what you don't want). In Midjourney, you use the --no parameter. In Stable Diffusion, there's a dedicated negative prompt input. In DALL-E 3, you describe what to avoid directly in your prompt text.
a beautiful portrait of a woman, soft lighting, 8K photography --no blurry, deformed hands, extra fingers, watermark, text, cropped, worst quality
The Universal Negative Prompt
Start with this base negative prompt for almost any portrait or photorealistic image. It eliminates the most common artifacts across Midjourney, Stable Diffusion, and similar tools:
blurry, out of focus, low quality, worst quality, jpeg artifacts, watermark, signature, text, logo, deformed, disfigured, extra limbs, extra fingers, missing fingers, fused fingers, mutated hands, poorly drawn hands, bad anatomy, bad proportions, cropped, poorly framed, ugly, lowres, grainy, noise, oversaturated, unnatural colors
Copy this as your starting point and customize it based on the specific issues you encounter in each session.
Category-Specific Negative Prompts
Portrait Photography
Portrait generation is the most common use case, and also the most prone to artifacts — especially around hands, eyes, and teeth.
deformed eyes, asymmetrical eyes, lazy eye, uneven pupils, bad teeth, missing teeth, floating teeth, extra teeth, cloned face, double face, ugly nose, bulbous nose, bad skin, plastic skin, smooth skin (too smooth), uncanny valley face, AI face, wax figure, mannequin look, body horror, extra fingers, mutated hands, six fingers
Landscape & Architecture
distorted perspective, impossible architecture, floating objects, inconsistent scale, blurry foreground, ugly sky, overexposed, flat lighting, low contrast, ugly colors, muddy colors, washed out, lens distortion (unless intended), fisheye (unless intended)
Digital Art & Fantasy
bad composition, poor anatomy, inconsistent lighting, mixed art styles (unless intended), low detail, amateurish, rough sketch, unfinished, scribbly, ugly color palette, muddy, flat, no depth
The "Before & After" Method
The most reliable way to learn what negative prompts to use is to run a generation without any negatives first, note exactly what's wrong, and then add precise negatives that address those specific issues.
a close-up portrait of a woman smiling, natural light, 8K --v 7
a close-up portrait of a woman smiling, natural light, 8K --no deformed teeth, uneven eyes, plastic skin, watermark, extra fingers --v 7
Common Mistakes With Negative Prompts
Mistake 1: Being Too Vague
Writing "bad quality" as a negative prompt is almost useless. The model doesn't have a clear concept of "bad quality" — it just knows statistical patterns. Be specific: "blurry, noisy, low-resolution, jpeg artifacts" are all concrete visual attributes the model understands.
Mistake 2: Contradicting Your Positive Prompt
If your positive prompt says "dark, moody, low-key lighting" and your negative prompt says "dark, shadow" — you'll get confused results. Always review your positive and negative prompts together to make sure they don't conflict.
Mistake 3: Overloading the Negative Prompt
More is not always better. Adding 50 negative terms can cause the model to overcorrect in unexpected ways. Start with 8–15 targeted negatives. Add more only when needed after reviewing results.
Mistake 4: Not Updating Negatives Per Session
Different subjects and styles have different failure modes. A fantasy illustration has different common artifacts than a photorealistic portrait. Build category-specific negative prompt templates (like the ones above) that you refine over time.
Negative Prompts in Stable Diffusion
In Stable Diffusion, negative prompts work even more powerfully because they directly affect the denoising process. The following is an industry-standard starting point for SD XL and SD 3.x:
(worst quality:2), (low quality:2), (normal quality:2), lowres, watermark, signature, username, artist name, bad anatomy, bad hands, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
Note the use of weighted terms like (worst quality:2) — Stable Diffusion allows you to weight individual terms, giving them more or less influence. A weight of 2 means double emphasis on avoiding that quality.
Using ZETRAX to Build Negative Prompts
The ZETRAX AI Prompt Builder includes pre-built negative prompt templates for every category — cinematic video, photography, digital art, and code prompts. Instead of remembering which negatives to use, simply select your style and the builder suggests the optimal negative configuration for that category.
This systematic approach ensures you never forget the basics, while leaving room to add your own custom negatives based on what you observe in each generation session.
Key Takeaways
- Negative prompts are as important as positive prompts — treat them with equal attention
- Be specific and visual in your negatives: "blurry hands" beats "bad quality"
- Use category-specific templates and update them as you discover new failure patterns
- Review positive and negative prompts together to ensure they don't contradict
- In Stable Diffusion, use weighted syntax
(term:1.5)for greater control - Start with 8–15 negatives and expand only when needed
Mastering negative prompts is one of the highest-leverage skills in AI image generation. Even small improvements here produce dramatically better outputs. Start building with ZETRAX and put these techniques into practice today.