Tip 12: Increase Weight

Have you noticed that the results generated by the Chinese figure prompt above seem to only have one Chinese face in the third image? The reason is that the weight of Chinese style is too low, so only one result is output.

In your practice, you will definitely encounter situations where AI ignores certain prompt words, such as mentioning a bird in the picture but it doesn't appear.

Why does this happen?

The reason is that Midjourney's prompt weights are gradually reduced in order (see Midjourney's official FAQ translated by me for details).

In our above prompt, we put Chinese Style at the end, so there are fewer results generated.

As for ignoring certain words, this reason remains unknown for now but it is likely due to weight issues.

There are many ways to increase weight.

The simplest method is to adjust the order, such as putting Chinese Style at the beginning (then generating results with squinty eyes style 🤣 - I did it twice and got better results on my second try).

product photography, Chinese Style,Joan of Arc, full body, hand painted, plastic, detailed, white background, studio lighting

Adjust the order, the effect is very different, so you do not believe in the so-called prompt templates, try more.

The second method is the Slider method.

You need to add , at the end of the prompt, then add the words you need to adjust the weight, then add , and finally add the weight value. For example, as in the example above, if I want to highlight Chinese style but don't want to adjust the order, I can change it to

product photography,Joan of Arc, full body, hand painted, plastic, detailed, white background, studio lighting, Chinese Style :: Chinese Style :: 0.5

Then you will find that the weights are too high, generating a graph that is not related to prompt (below left), so we can add the weights of Joan of Arc to the original (below right):

product photography,Joan of Arc, full body, hand painted, plastic, detailed, white background, studio lighting, Chinese Style :: Joan of Arc:: 0.25 style of Chinese:: 0.5

You can see that after adjusting the parameters, 2 of the 4 resulting images have a strong Chinese style:

But frankly speaking, this method, is not particularly suitable for this case, adjust the main content, or order, ROI will be higher, this method is more suitable for miss drop a prompt word when used, such as the official case is this:

several sad cats watch a crying woman in a home-office environment with laptop

Generated pictures (four pictures on the left), the cats are looking at you (you may be the crying woman 😂), if you use the function of slider (four on the right), crying woman will appear in the picture:

several sad cats watch a crying woman in a home-office environment with laptop:: several sad cats::0.5 crying woman::0.6 laptop::0.7

Additionally, it should be noted that the available parameters for the slider are 0.25, 0.5, 0.6, and 0.7. This method can also be used to decrease weight with intervals of -0.7, -0.6, -0.5, and -0.25.

If you have used the slider but still cannot see the expected image, there is another trick called cowbell. To be honest, this does not comply with Midjourney's official best practices, but it is indeed a feasible method. Take the following prompt as an example: in the generated images, the woman does not look shy (the left four pictures).

A songbird sits on the shoulder of a shy blonde woman dressed in white.

At this point you can repeat and repeat and repeat (I think that's why the method is called cowbell):

A songbird sits on the shoulder shoulder shoulder shoulder shoulder shoulder of a shy shy shy shy shy shy shy blonde woman dressed in white.