Easy to Follow, From Beginner to Advanced, Guide For Midjourney AI Image Generation. Part 4 - Multi Prompts With Weights

Before you start with Midjourney please understand the Guidelines and Terms of Service and use this tool responsibly

Quick note: If you missed previous guides, check them out before you continue with this one.

Part 1: https://peakd.com/hive-122472/@awesomeintrigue/easy-to-follow-from-beginner-to-advanced-guide-for-midjourney-ai-image-generation-part-1-lets-get-started

Part 2: https://peakd.com/hive-122472/@awesomeintrigue/easy-to-follow-from-beginner-to-advanced-guide-for-midjourney-ai-image-generation-part-2-adding-some-style

Part 3: https://peakd.com/hive-122472/@awesomeintrigue/easy-to-follow-from-beginner

Prompt Building Tool: https://peakd.com/hive-122472/@awesomeintrigue/midjourney-prompt-tool-easy-way

So far I think we covered all the basics. How to write a simple text prompt, image prompts, and how to use parameters. If you missed any of those go check out earlier posts that are listed above.

There is one more fundamental thing that we need to look at more precisely because that will be directly tied to Multi Prompting and understanding Weights. And to grasp fully how those work we need to answer one question.

How does Midjourney actually perceive our prompts?

Just to clarify. There is no exact consensus as to how the whole process of interpreting prompts works in Midjourney so these are just my conclusions from my own experiments and what I've seen from other people.

As we already learned, the structure of the prompt: is an image prompt, text prompt, and parameters. In part 3 we discussed how image prompts can have different uses depending on their position in the prompt. Text prompts on the other hand are much simpler, and straightforward.

Before we go into multi-prompts and weights I want to bust some misconceptions about the importance of punctuation and separating text prompts with commas. This isn't exactly true. But don't trust me on my word, let me show you.

We are going to use this prompt for comparison: /imagine elegant woman, samurai warrior, greenish-black topknot hair, red shiny armor, daimyo --seed 420

We are going to use the same see parameter with all of these examples to get consistent results that we can compare, and so that you can try it out yourself. Anyway, the result from the first prompt is this:

Fair enough, we got what we want. Let's compare it to this weird looking prompt: /imagine "elegant woman" (samurai warrior), greenish-black topknot hair "red shiny armor, daimyo" --seed 420

No surprises here, we got what we wanted, and it's pretty similar. The first images from both prompts look almost the same but other images differ a bit more.

Let's now try this one without any punctuation whatsoever: /imagine elegant woman samurai warrior greenish-black topknot hair red shiny armor daimyo --seed 420

Unlike our second prompt, this one almost exactly matches the first prompt. Changes are only in small details. If you want to compare them side by side open this link (first prompt is on the left, and the third one is on the right).

You can play with random punctuation signs and commas in different places if you want to compare more results but this is my conclusion.

The prompt with commas and the one without are almost the same, while the one with a bunch of punctuation signs didn't really affect the results too much. Midjourney seems to be looking at the whole prompt and is trying to recognize patterns that make sense to it. The concept of black hair or topknot hairstyle is far more important than if there is a comma, parathesis, or period. Even though these signs don't exactly matter to Midjourney, it will process their existence in the prompt, adding a bit of noise in the overall output, hence different details in the image.

---

I believe that it's still useful to use punctuation even though it may not matter to Midjourney. Well-structured prompts are easy to read, and they help you identify any mistakes or typos in your prompts (especially when you write a really long prompt). In other words, write prompts with punctuation for yourself.

We can safely say that Midjourney sees spaces between words, right? Well, kinda.

So we used the same prompt, /imagine elegant woman, samurai warrior, greenish-black topknot hair, red shiny armor, daimyo --seed 420 without spaces or to be more specific this prompt: /imagine elegantwomansamuraiwarriorgreenish-blacktopknothairredshinyarmordaimyo --seed 420

Funnily enough, we got pretty close to our initial prompt, haha. Sure, we lost some important details like that it should be a woman but we got a samurai. Also, it seems that the AI didn't know what we wanted it to do with the colors so we got green and black all over the place but the overall concept isn't lost.

So what does Midjourney actually see?

To answer this let's write this prompt in a so-called multi-prompt way. To truly separate different concepts for Midjourney instead of using commas or periods or whatever, we are going to use the double colon (::) between our prompts. For example, this is how our original prompt with these separators looks like:

/imagine elegant woman:: samurai warrior:: greenish-black topknot hair:: red shiny armor:: daimyo:: --seed 420

Would you look at that? This work looks so much more polished and we got pretty much what we wanted. Separating concepts into multiple prompts by using (::) seems to be the way most people thought commas worked. But that's not all, we can go even further and add weights.

Weights

Let's take a look at our last prompt again: elegant woman:: samurai warrior:: greenish-black topknot hair:: red shiny armor:: daimyo:: --seed 420

As you can see we separated this prompt into 5 parts which means that Midjourney will try to compose an image with those 5 concepts focusing on them equally. What we can do here is implement weights. After each (::) you can add a number that will represent a weight. By default, every weight is set to 1. So we can write our prompt like this to get the same result:

/imagine elegant woman::1 samurai warrior::1 greenish-black topknot hair::1 red shiny armor::1 daimyo::1 --seed 420

We can also look at weights as percentages. Since we have 5 prompts with equal weighting, they are split into 20% relevancy each. In all of our examples so far I haven't really got any satisfying results for the red shiny armor part. So let's change the value of that prompt to 1.5, meaning that its relevancy goes up to 30% while other prompts go down to 17.5%.

With a slight change, our prompt looks like this: /imagine elegant woman::1 samurai warrior::1 greenish-black topknot hair::1 red shiny armor::1.5 daimyo::1 --seed 420

Much better! The armor is indeed redder and shinier. We can go even further. Let's say we want more emphasis on the samurai part but also keep the importance of the armor texture. So I'm gonna add 3 to both samurai warrior and red shiny armor prompts.

Our prompt now looks like this: /imagine elegant woman::1 samurai warrior::3 greenish-black topknot hair::1 red shiny armor::3 daimyo::1 --seed 420

As you can see, with a bigger emphasis on samurai and armor, we are starting to see how other prompts, such as hair and samurai actually being a woman start to matter less. In imagine 1, Midjourney didn't even use those concepts.

Weights can be such a powerful tool for fine-tuning images but they can also completely screw up our outputs if we forget about the balance.

Prompt for this image was this: /imagine elegant woman::30 samurai warrior::1 greenish-black topknot hair::50 red shiny armor::1 daimyo::1

Just to show an example I exaggerated prompts "elegant woman" and "greenish-black topknot hair" with absurdly heavy weights. Because of those absurd weights, prompts with the weight of 1 have only about 1.2% relevancy for the output and Midjourney decided to not even bother with them so it basically just used those two prompts with much bigger weights.

In conclusion

I'm not saying that I have the right or wrong interpretation about how you should be writing your prompt but these are my observations.

If you want a more creative and free-form image then write everything in one prompt. But if you prefer something specific that looks exactly how you imagined it, you should approach it with multi-prompting.

Both ways of prompting need to be explored even further, but to me, this distinction seems obvious. If anyone thinks I'm off with this interpretation please correct me in the comments.

If you have more questions or need clarification please reach out.

And as always, thanks for reading.

IF you need help starting here at Hive, I highly recommend you to hop on The Terminal discord server, here's an invite: https://discord.gg/Wv5yQwyF