You can use negative prompts to specify what you don't want in your Midjourney generations.
Here's an example of where you could use a negative prompt:
When you use the word "painting
" in your prompts you'll get photographs of paintings mounted on walls, sometimes with additions like a frame and furniture.
Let's say you just want the painting by itself. You would use negative prompting to get rid of all of the extra details:
Original

modern abstract painting --seed 1
w/ Negative Prompts

modern abstract painting --seed 1 --no frame furniture wall
Protip: Put all your negative prompts after a single --no
, there's no need to write a --no
for every individual thing
Examples of effective negative prompts
Use negative prompts to get rid of hands
It's not exactly a secret Midjourney can't do hands very well (yet). If you're getting too many weird hands, why not just hide them?
Original

blonde girl leaning on table and smiling at camera, detailed facial features, detailed eyes, polaroid, 1990
w/ Negative Prompts

blonde girl leaning on table and smiling at camera, detailed facial features, detailed eyes, polaroid, 1990 --no hands
Use Negative Prompts to get rid of text
Text is another one of those things that Midjourney just doesn't do well. You can get rid of text with negative prompts.
The more elements you specify the better. I used "--no text font letters
", which works much better than just "--no text":
Original

chicken logo, style of Milton Glaser
w/ Negative Prompts

chicken logo, style of Milton Glaser --no text font letters
You can see in this example, the negative prompt not only rid of the text, but it also made the images closer to the style I wanted (Milton Glaser, designer of the New York subway map) Midjourney was able to emulate the style better when it didn't have to focus on composing a logo with text.
You can observe a similar effect here:
Original

chicken logo, style of Egon Schiele
w/ Negative Prompts

chicken logo, style of Egon Schiele --no text font letters
When negative prompted to not include text, the end result is much closer to the artists' actual style.
This is a good way to transition into:
Use negative prompts to get the style you want
Here's a prompt I tried. While the original results are pretty, they're not what I wanted at all.
Original

anime stormtrooper by studio ghibli
w/ Negative Prompts

anime stormtrooper by studio ghibli --no 3d render realistic
My hypothesis was that there were just too many 3D & photorealistic stormtroopers in the training data. It completely overpowered the rest of the prompt!
By specifying that I didn't want 3D in my negative prompt, I was able to get results much closer to what I was looking for.
Negative prompt while keeping the same Seed
Let's say you already have an image you like. You want to add a negative prompt to it. But you don't want to re-roll, because that would create different images.
You can re-create that exact image with a negative prompt, as long as you know the seed
.
Every image has a seed, and images with the same seed are similar to one another. Here's how to find the seed of any image.
After you find the seed you can do very cool things like this:
Original

light watercolor, interior of a cozy cafe, bright, white background, few details, dreamy, Studio Ghibli --seed 3817455947
Same Seed w/ Negative Prompts

light watercolor, interior of a cozy cafe, bright, white background, few details, dreamy, Studio Ghibli --seed 3817455947 --no people person
Alternate method: use negative image weights
Apart from using Negative Prompts, you can also use negative weights to specify what you don't want in your generations.
A little bit more on weights:
Weights are used to emphasize and de-emphasize different parts of a prompt.
Words with positive weights are emphasized while words with negative weights are de-emphasized.
The actual numbers don't matter: what matters is their proportion to each other. Words have a default weight of 1.
I prefer using weights over the "--no" command because they gives you fine-tuned control of every part of the prompt.
Here's how I would write the above example with weights:
anime by studio ghibli::1.4 stormtrooper::1.2 3d render realistic::-1.5
Follow up every part of the prompt with ::INSERT-WEIGHT-HERE.
Here are the results:

anime by studio ghibli::1.4 stormtrooper::1.2
3d render realistic::-1.5

anime by studio ghibli::1.5 stormtrooper::1
3d render realistic::-1.5
wheigt can be which number? any number? remember, those you addressed, suppose to not know this stuff…
Hey Sorin, I have updated the article!
I tried this and it doesn’t generate the same picture at all. I used the — seed# – gave me errors I had to remove all spaces before seed >> –seed#.
Hey Megan. This guide was written at the time of Midjourney V4. If you want the exact images, try with “–v 4”