¶ Prompt Weighting
¶ What is prompt weighting?
Sometimes the AI will ignore parts of your prompt. Prompt weighting is a powerful tool that allows you to add weight or importance to a word or words in your prompt that may otherwise be ignored.
Here's an example where I wanted to create a mosaic design:
 |
 |
| Prompt: giant pint robot in the ocean, surrounded by ocean waves, mosaic art |
Prompt: giant pink robot in the ocean, surrounded by ocean waves, (mosaic art:1.3) |
In the example on the left, the AI gave the robot a subtle mosaic texture, but there is no mosaic in the background. Once I added prompt weighting, as shown in the example on the right, the AI created a true mosaic design.
¶ How does prompt weighting work?
Put the word or words you want to weight inside a parenthesis, followed by a colon and a number, like this:
Beautiful woman in a rose garden, wearing a (purple lace dress:1.2), holding a (basket of red roses:1.3)
The greater the number, the greater the weight. You can go all the way up to 1.9, but no higher! If you want to reduce the weight of a term, you can use a smaller number, like this:
Beautiful woman in a rose garden, at night, (eerie:0.8), insane details, realistic, (dark:1.4)
In this case, I wanted just a hint of eeriness, but more than none.
¶ Is there another way to do prompt weighting?
Yes. You can also increase or decrease the weight of a term by adding plusses or minuses, like this:
Beautiful woman in a rose garden, wearing a (purple lace dress)++, at night, eerie--, dark++++
+ is equivalent to 1.1; however, ++ is 1.1 x 1.1 or 1.21, which is not exactly the same as 1.2
Similarly, – is 0.9 x 0.9 which is 0.81
Ergo, the plus/minus method will give slightly different results than the numeric method, UNLESS you used multiple decimal places.
¶ Can I use multiple decimal places?
Yes.
Beautiful woman in a rose garden, wearing a (purple lace dress:1.21), at night, (eerie:0.81), (dark:1.4641)
Will give the exact same result as the plus/minus example I used above.
¶ What about parentheses and square brackets?
Putting something in parentheses has the same effect as weighting 1.1. Putting something in square brackets has the same effect as weighting 0.9.
| These all add .1 | These all subtract .1 |
|
Woman in a (purple dress:1.1) Woman in a (purple dress)+ Woman in a (purple dress) |
Woman in a (purple dress:0.9) Woman in a (purple dress)- Woman in a [purple dress] |
More than 2 nested parentheses will be ignored. More than 1 set of square brackets will be ignored.
¶ What are nested parentheses?
Nested parentheses are parentheses inside parentheses, like this ((purple dress)). This is not a recommended way to prompt. However, it can be useful for separating features of two characters, like this:
2 girls A and B, (A girl has blonde hair, (a red dress:1.4), (pink bow in her hair:1.2)), (B girl has black hair, (a yellow shirt and denim shorts:1.3)), hugging
In this case, the nested parentheses help to keep the description of the first character separate from that of the second character.
¶ Does prompt weighting work for negative prompts?
Yes.
¶ Which models don't work with prompt weighting?
Flux Schnell
Stable Diffusion 3.5 Medium
Stable Diffusion 3.5 Large
Prompt weighting will not be effective with these models.