> ## Content Index
> Fetch the complete content index at: https://www.hyperank.ai/blog/llms.txt
> Use this file to discover other available public pages before exploring further.

# What Is a Negative Prompt? Examples for AI Image Generators
- URL: https://www.hyperank.ai/blog/negative-prompts/
- Published: 2026-11-03T07:30:13.000Z
- Updated: 2026-11-03T07:30:13.000Z
- Description: Discover what a negative prompt is and explore examples for AI image generators to enhance your creative projects effectively.
- Author: Hyperank AI

*negative prompt | Updated October 2026 | 9 min read | hyperank Editorial Team*

**A negative prompt is an instruction that tells an AI image generator what you don't want in the image, such as "blurry, extra fingers, text, watermark," so the model steers away from those features.** In Stable Diffusion tools it's usually a separate field next to the main prompt, and in Midjourney it's the --no parameter. Not every tool supports negative prompts, and for text models like ChatGPT and Claude, telling the model what to do instead usually works better than listing what to avoid. This guide, What Is a Negative Prompt? Examples for AI Image Generators, explains how negative prompts work, which tools support them, common examples you can copy, and how to write negative instructions for text models.

> A good prompt describes the picture you want. A good negative prompt removes the things that keep showing up uninvited.

## What Is a Negative Prompt?

A negative prompt is a list of words or phrases describing elements, styles or flaws you want an AI image generator to avoid. It works alongside your normal (positive) prompt: the positive prompt says what to create, and the negative prompt says what to keep out.

| Prompt type     | Example                                                                | Purpose                      |
| --------------- | ---------------------------------------------------------------------- | ---------------------------- |
| Positive prompt | "Portrait of a woman reading in a sunlit cafe, soft light, 35mm photo" | Describes the image you want |
| Negative prompt | "blurry, distorted hands, text, watermark, oversaturated"              | Lists features to avoid      |

- **Quality fixes:** Reduce common flaws like blur, artifacts or distorted anatomy.
- **Content control:** Keep out objects, people, text or logos.
- **Style control:** Avoid styles you don't want, such as cartoonish or overly dark.

### Why negative prompts exist

Image models often add details you never asked for, because those details are common in their training data. Writing "no people" in the main prompt can backfire, since the model may latch onto the word "people." A separate negative prompt gives the model a cleaner signal of what to avoid.

## How Negative Prompts Work

In diffusion models such as Stable Diffusion, an image is created by gradually removing noise, guided by your prompt. A negative prompt gives the model a second direction: at each step it moves toward the positive prompt and away from the negative one. Hugging Face's [Diffusers documentation](https://huggingface.co/docs/diffusers/en/using-diffusers/weighted%5Fprompts?ref=hyperank.ai) exposes this directly as a negative\_prompt setting.

- **Guidance:** The model compares predictions for the positive and negative prompts and pushes the image toward the positive.
- **Not a filter:** A negative prompt reduces the chance of a feature; it doesn't guarantee removal.
- **Interaction with settings:** Guidance strength (often called CFG scale) affects how strongly both prompts are followed.
- **Research:** A [2024 study on negative prompts](https://arxiv.org/html/2406.02965v1?ref=hyperank.ai) examined when and how they take effect during image generation.

### Why short and specific works best

Long, kitchen-sink negative prompts can dull an image or conflict with your positive prompt. Start with the specific problems you see in your outputs, add them to the negative prompt, and regenerate.

### A simple testing method

- **Fix the seed:** Where your tool allows it, keep the same seed so changes come from the prompt, not randomness.
- **Change one thing:** Add or remove a single negative term per test.
- **Compare side by side:** Generate a few images with and without the term before deciding it helps.

## Which Image Tools Support Negative Prompts

Support varies by tool. Stable Diffusion interfaces and many tools built on open models include a dedicated negative prompt field; Midjourney uses a parameter; and some popular generators don't offer a separate field at all.

| Tool                                                 | Negative prompt support        | How to use it                                             |
| ---------------------------------------------------- | ------------------------------ | --------------------------------------------------------- |
| Stable Diffusion interfaces (AUTOMATIC1111, ComfyUI) | Yes, dedicated field           | Type unwanted features in the negative prompt box         |
| Hugging Face Diffusers                               | Yes, negative\_prompt argument | Pass a string to the pipeline                             |
| Midjourney                                           | Yes, --no parameter            | Add --no followed by items, for example --no text, people |
| Leonardo AI                                          | Yes, negative prompt option    | Enable and fill the negative prompt box                   |
| ChatGPT image generation                             | No separate field              | Describe what you want clearly; ask for edits afterwards  |

- **Midjourney:** [Midjourney's documentation](https://docs.midjourney.com/docs/no?ref=hyperank.ai) describes --no as a way to tell the model what you don't want, and notes it works like giving that part of a prompt a negative weight.
- **Check each tool:** Features change often, so confirm in the tool's current help pages before relying on a negative prompt.
- **No field?** Rephrase the positive prompt to describe the scene precisely, such as "empty beach at dawn" instead of "beach, no people."

## Common Negative Prompt Examples

These negative prompts address the most common problems in AI images. Use only the ones that match issues you're actually seeing, rather than pasting them all.

| Goal              | Example negative prompt                                    |
| ----------------- | ---------------------------------------------------------- |
| General quality   | blurry, low quality, jpeg artifacts, noisy, out of focus   |
| Anatomy           | distorted hands, extra fingers, extra limbs, deformed face |
| Text and branding | text, watermark, logo, signature, caption                  |
| Composition       | cropped, cut off, cluttered background, duplicate objects  |
| Style             | cartoon, illustration, 3D render (when you want a photo)   |
| Color and light   | oversaturated, overexposed, too dark, harsh shadows        |
| Product shots     | reflections, fingerprints, props, busy background          |

### Weighting negative terms

Some Stable Diffusion interfaces let you weight terms, for example by wrapping a word in parentheses or adding a number, to make a negative term stronger or weaker. Syntax varies by interface, so check its documentation, and use weights sparingly, since heavy weights can distort the whole image.

### Example pairs

- **Product photo:** Positive "white sneaker on a plain light gray background, studio lighting, high detail"; negative "text, logo, shadows, props, reflections."
- **Landscape:** Positive "mountain lake at sunrise, mist, wide angle"; negative "people, buildings, boats, oversaturated."
- **Portrait:** Positive "headshot of a smiling man in a navy jacket, natural window light"; negative "distorted face, extra fingers, harsh shadows, blurry."

For writing the positive side, see our guide on [how to write AI image prompts](https://www.hyperank.ai/blog/ai-image-prompts).

**Key Takeaway:** Fix one problem at a time. Add a term to the negative prompt only after you've seen that problem in your results.

## Writing Negative Instructions for Text Models

Text models like ChatGPT and Claude don't have a separate negative prompt field, and listing what not to do is often less effective than stating what to do. Anthropic's [prompting best practices](https://docs.claude.com/en/docs/build-with-claude/prompt-engineering/claude-4-best-practices?ref=hyperank.ai) recommend you "tell Claude what to do instead of what not to do," giving the example of replacing "Do not use markdown in your response" with "Your response should be composed of smoothly flowing prose paragraphs."

| Instead of                 | Try                                                              |
| -------------------------- | ---------------------------------------------------------------- |
| "Don't be too long."       | "Answer in no more than 150 words."                              |
| "Don't use jargon."        | "Write for a reader with no technical background."               |
| "Don't make things up."    | "Use only the facts below; write \[CHECK\] where you're unsure." |
| "Don't use bullet points." | "Write in short paragraphs of two to three sentences."           |

- **When negatives still help:** Hard rules, such as "Never include customer names," are worth stating explicitly, ideally alongside the positive alternative.
- **Keep them few:** A long list of "don'ts" can distract from the main task.
- **Test it:** Run the prompt a few times and check whether the rule is followed consistently.

### Tools to help

The free [AI prompt generator](https://www.hyperank.ai/tools/ai-visibility/ai-prompt-generator?ref=hyperank.ai) helps turn vague requests into clear, positive instructions. If you use AI to create content about your brand, hyperank also tracks what AI engines such as ChatGPT, Perplexity and Gemini say about you, which is a separate but related question from how you prompt them.

> With images, negative prompts remove clutter. With text, positive instructions prevent it.

## Negative Prompt FAQ

The most common questions are whether every tool supports negative prompts, whether they guarantee removal, and how many terms to use. Short answers are below.

- **Every tool?** No; check each tool's documentation.
- **Guaranteed?** No; they reduce, not eliminate, unwanted features.
- **How many terms?** As few as solve the problem you see.

## Conclusion

A negative prompt tells an AI image generator what to leave out, steering it away from flaws, objects and styles you don't want. Use the dedicated field in Stable Diffusion tools or Midjourney's --no parameter, keep negatives short and specific, and for text models, state what you want instead of what to avoid.

### Key Takeaways

- **Definition:** A list of features you want the image model to avoid.
- **How it works:** Guides diffusion away from the negative prompt at each step.
- **Support:** Stable Diffusion tools, Diffusers, Midjourney (--no) and Leonardo AI.
- **Examples:** Quality, anatomy, text, composition and style fixes.
- **Text models:** Prefer positive instructions.

Write clearer prompts with the free AI prompt generator.

## FAQ

### What is a negative prompt, and what are examples for AI image generators?

A negative prompt tells an AI image generator what to avoid, alongside the main prompt that describes what to create. Common examples include "blurry, low quality" for quality, "extra fingers, distorted hands" for anatomy, "text, watermark, logo" for branding, and "cartoon, illustration" when you want a photo. Stable Diffusion tools use a dedicated field, and Midjourney uses the --no parameter.

### Does Midjourney have negative prompts?

Yes, through the --no parameter, for example "--no text, people."

### Does ChatGPT support negative prompts for images?

There's no separate negative prompt field. Describe the image precisely and request edits if unwanted elements appear.

### Do negative prompts always work?

No. They reduce the likelihood of a feature appearing but don't guarantee it's removed.

### Should I use a long negative prompt?

Usually not. Short, specific negatives targeting problems you've seen work better.

### How do I write negative instructions for ChatGPT or Claude?

State what you want instead, for example "answer in 150 words" rather than "don't be long," and keep any hard "never" rules few and explicit.

*Methodology: This article draws on Hugging Face Diffusers documentation, Midjourney's --no parameter documentation, a 2024 research paper on negative prompts, and Anthropic's prompting best practices. Tool features change often; confirm current support in each tool's help pages.*