- How negative prompts work in each tool
- The Midjourney rule almost everyone gets wrong
- Semantic negative prompts for Gemini and ChatGPT
- Negative prompts for product photos
- Negative prompts for people and headshots
- Negative prompts for text, logos and graphics
- Negative prompts for backgrounds and scenes
- Quick-fix lists you can bolt on
- How to get better results from negative prompts
- My verdict
- Frequently asked questions
A negative prompt tells an AI image tool what to leave out of a picture. In Midjourney you add --no followed by a comma-separated list, like --no text, watermark, clutter. In Gemini and ChatGPT there's no separate box, so you describe the scene you do want instead, which Google calls a semantic negative prompt.
Most bad AI images aren't wrong because of what's in the prompt. They're wrong because of what crept in uninvited: a stray logo, a sixth finger, a messy background, text that looks like it was typed by a cat. Negative prompts are how you shut the door on those extras.
The catch is that every tool handles them differently, and the way most people write them ("please don't add any people") often backfires. Below you'll find how each type of tool expects them, then copy-paste negative prompt sets for the jobs small businesses use AI images for most.
How negative prompts work in each tool
| Type of tool | How you exclude things | What to watch for |
|---|---|---|
| Midjourney | Add --no at the end, then a comma-separated list | Each word is read on its own, so avoid two-word phrases that change meaning |
| Stable Diffusion apps with a negative field | Type the list into the separate negative prompt box | Keep it short and specific to the image |
| Gemini image generation (Nano Banana) | Describe the scene positively | Saying 'no cars' can still put cars in your head and the picture |
| ChatGPT images | Describe what you want, then ask for edits in follow-ups | Fix one thing per follow-up message |
The Midjourney rule almost everyone gets wrong
Midjourney's own docs are blunt about this. Writing "without any fruit" or "please don't add fruit" in your main prompt is listed as a bad example. The good version puts the exclusions after --no: still life gouache painting --no fruit, apple, pear.
The bit that trips people up is how it reads that list. Midjourney's moderation looks at every word in --no separately. So --no modern clothing is read as "no modern" and "no clothing", which can set off a content warning. Their advice is to describe the clothing you do want in the main prompt instead. Under the bonnet, --no works like giving that part of a multi-prompt a weight of minus 0.5, which is why it nudges rather than bans.
Semantic negative prompts for Gemini and ChatGPT
Google's image generation guide recommends what it calls semantic negative prompts. Rather than "no cars", you write "an empty, deserted street with no signs of traffic". You're painting the absence instead of naming the thing. It sounds fussy, but it works because you're giving the model a clear picture to aim at.
Here's the same idea translated into business jobs. Each prompt below comes in two versions where it matters: a Midjourney line with --no, and a positive rewrite for Gemini or ChatGPT.
Negative prompts for product photos
Clean product shot (Midjourney)
Copy this prompt
studio product photograph of a [product] on a plain [colour] background, soft even lighting, sharp focus --no text, logo, watermark, props, clutter, reflections, shadows
Clean product shot (Gemini or ChatGPT)
Copy this prompt
A studio product photograph of a [product] on a plain [colour] background. The surface is empty apart from the product, with soft, even light and no lettering anywhere in the frame.
Lifestyle product shot
Copy this prompt
A [product] on a tidy kitchen worktop in morning light. The worktop is clear except for one mug and a folded linen cloth. No people in the frame, nothing branded in view.
Flat lay for social posts
Copy this prompt
Top-down flat lay of [items] on a pale wooden table, evenly spaced with clear gaps between them, a calm uncluttered look, no handwriting or printed words.
Negative prompts for people and headshots
Professional headshot (Midjourney)
Copy this prompt
professional headshot of a [age] [description] business owner, neutral grey backdrop, natural smile, soft window light --no jewellery, hat, sunglasses, busy background, text
Natural-looking hands
Copy this prompt
A [person] holding a coffee cup with both hands relaxed around it, five fingers clearly visible on each hand, simple realistic pose.
Team photo without the stock-photo look
Copy this prompt
A small team of four colleagues chatting around a table in a bright office, candid and relaxed, nobody looking at the camera, plain walls, no posters or signage.
Customer avatar illustration
Copy this prompt
Flat illustration of a [description] customer working on a laptop at home, simple shapes, two-colour palette, a clean background with no extra objects.
Negative prompts for text, logos and graphics
Background for a quote graphic (Midjourney)
Copy this prompt
soft abstract gradient background in [brand colours], gentle grain, lots of empty space in the centre --no text, letters, numbers, symbols, faces
Logo concept exploration
Copy this prompt
Simple logo concept for a [type of business] called [name], one flat icon beside the name, single colour on white, nothing else on the canvas.
Banner with room for your headline
Copy this prompt
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
A wide website banner showing [scene], with the left third left as calm empty sky so text can sit there later. The image itself contains no words.
Infographic icon set
Copy this prompt
A set of six matching line icons for [topics], same stroke weight, evenly spaced on white, no labels under the icons.
Negative prompts for backgrounds and scenes
Empty high street
Copy this prompt
A quiet high street at dawn, empty pavements and a deserted road with no traffic, shopfronts with blank signs, soft blue light.
Clean workspace
Copy this prompt
A minimal home office desk with only a laptop and a notebook on it, cables hidden, a plain wall behind, calm natural light.
Outdoor scene without crowds (Midjourney)
Copy this prompt
wide shot of a sandy beach at sunrise, gentle waves, pastel sky --no people, boats, buildings, litter, birds
Room shot for an interior business
Copy this prompt
A bright living room styled for a [style] interior, every surface tidy, no clutter, no visible brand names, photographed straight on at eye level.
Quick-fix lists you can bolt on
Common clean-up list (Midjourney)
Copy this prompt
--no text, watermark, signature, logo, border, frame, blur
Fix extra objects (follow-up in ChatGPT or Gemini)
Copy this prompt
Keep everything the same, but remove the objects in the background so the wall behind the subject is plain.
Fix a busy composition (follow-up)
Copy this prompt
Keep the subject and lighting exactly as they're. Simplify the background to a single soft colour.
Fix unwanted text (follow-up)
Copy this prompt
Keep the image the same but remove every piece of writing, including on signs, packaging and screens.
How to get better results from negative prompts
- Exclude what you've seen, not what you fear. Run the prompt once, look at what went wrong, then add only those items to your list.
- Use single words in Midjourney. Because each word in --no is read separately, pick words that are safe on their own.
- Describe the gap positively everywhere else. "A clear wall" beats "no clutter" in Gemini and ChatGPT, exactly as Google's guide suggests.
- Fix one thing per follow-up. Google's guide recommends iterating with small changes, and that's where chat-based tools shine.
- Give the purpose. Saying the image is for a product page or a LinkedIn banner helps the model decide what doesn't belong.
If you're building a library of image prompts for your brand, my Gemini prompts for business owners pair nicely with these, and the Kling AI prompts cover the same idea for video.
My verdict
Negative prompts are worth learning, but they're a clean-up tool, not a starting point. Write a clear, specific prompt first, look at the result, then exclude the two or three things that keep sneaking in. In Midjourney use --no with single words; everywhere else, describe the empty space you want. That one habit will save you more rerolls than any clever list.
Frequently asked questions
What is a negative prompt in AI image generation?
It's an instruction telling the tool what to leave out of an image, such as text, watermarks, extra people or clutter. Some tools have a separate field or parameter for it, others expect you to describe the scene positively.
How do I write a negative prompt in Midjourney?
Add --no at the end of your prompt, followed by the things you don't want, separated by commas, for example --no text, logo, people.
Does ChatGPT have a negative prompt box?
No. Describe what you want to see, then use follow-up messages to remove anything that slips in, one change at a time.
Why does Midjourney give me a warning when I use --no?
Its moderation reads each word in --no on its own, so a phrase like 'modern clothing' becomes 'no modern' and 'no clothing'. Describe the clothing you do want in the main prompt instead.
What is a semantic negative prompt?
It's Google's term for describing the absence positively. Instead of 'no cars', you write 'an empty, deserted street with no signs of traffic'.
