Multi Prompts: How to Get Separate Concepts in Your Image

In this newsletter, read about:

• 🕵️‍♀️ Multi Prompts in Midjourney

• 🗞 News and Top Reads

• 📌 AI Art Tutorial: Testing AI Art Generators

• 🎨 Featured Artist: Dehiscence

• 🖼 AI-Assisted Artwork of the Week

• 🤓 A Comprehensive Midjourney Guide

🕵️‍♀️ Multi Prompts in Midjourney

In the ever-expanding array of AI image generators, Midjourney stands out as an exceptional tool that pushes the boundaries of creativity. However, we all understand the challenges of obtaining an image that precisely depicts what you are seeking. In previous editions of this newsletter and my Midjourney guide, I explored various features of Midjourney that give users greater control over the resulting image.

Today, I would like to delve into multi prompts, another Midjourney feature that enables the depiction of distinct concepts within a single image, and moreover, assigning different levels of importance to these separate concepts. Let’s delve deeper into the mechanics of this feature, exploring how it works and providing more examples to showcase its versatility.

Separating Concepts

To instruct Midjourney Bot to consider two or more separate concepts individually, use a double semicolon (::) as a separator.

For instance, your results will differ dramatically when you're requesting “hot dog” vs. “hot:: dog”.

Similarly, if you prompt for “still life painting”, you’ll get the classic paintings with fruits and vases on a table, while “still:: life:: painting” will yield a variety of artworks depicting calm and peaceful life.

Assigning Different Weights

A double colon (::) functions not only as a separator but also as a means to assign varying levels of importance to different parts of the prompt.

Let’s say we start with the prompt, where we separate the two concepts of “fire” and “water”, but keep the relative weights equal.

If we want more water in the output image, we can simply assign this part a higher weight.

Changing the prompt to “fire:: water::2” amplifies the significance of the word water, making it twice as important as the word fire. As a result, the images generated are predominantly focused on water, with fire serving as a complementary element. This is clearly evident in the exemplary images provided above.

Here are a few things to remember when using prompt weights in Midjourney:

• Non-specified weights default to 1.

• The latest versions (v4 and later) accept decimal values for weights.

• Weights are normalized, implying that:

  • “fire:: water”, “fire::1 water”, “fire::2 water::2”, and “fire::50 water::50” are all the same;

  • “fire:: water::2”, “fire::2 water::4”, and “fire::50 water::100” are all the same.

• Negative weights can be used to reduce the impact of certain elements or remove them entirely.

• The sum of all weights used in the prompt must be a positive number.

To gain a better understanding of how this concept operates in practice, let's explore the idea of negative weights using an example.

Removing Unwanted Elements

We’ll start with a basic prompt, requesting a landscape in the surrealist style.

Now, let's say we want to reduce the presence of water in the image. We can accomplish this by assigning a negative weight to the water concept. However, as we are aware, the total sum of all weights should be positive. This implies that we need to select one or more additional concepts and assign them a combined weight of 0.5 or greater.

In this particular case, I will simply request an increase in the number of trees instead of water.

Alternatively, you can experiment with colors by, for instance, asking for a greater amount of blue and a lesser amount of red.

Note that weighing part of a multi prompt to "-.5” is the same as using a --no parameter with the corresponding word. Specifically, “a stunning landscape in a surrealist style with trees:: water::-.5” is the same as “a stunning landscape in a surrealist style with trees:: --no water”.

Thus, if the "--no" parameter is not sufficient in your case, you can utilize a negative weight below "-0.5" (e.g., "-1").

Now, the output image grid no longer includes any water whatsoever, and not even blue color resembling water.

Summing Up: More Control with Multi Promps and Weights

As you can see, multi-prompts can be very helpful in exerting more control over the image output. However, keep in mind that image generation is a stochastic process and text weights may not always function as intended, particularly when dealing with complex prompts involving multiple distinct concepts.

Now that you are familiar with the fundamental principles of multi-prompts, you can begin experimenting with them, bearing in mind that often, you will need to re-roll and refine your prompt multiple times before achieving the desired outcome.

Happy prompting!

🗞 News and Top Reads

• Meta introduced CM3Leon – a state-of-the-art generative model for text and images.

  • The authors claim that this model can generate coherent images given prompt text with potentially highly compositional structure.

  • CM3Leon also supports text-guided image editing (e.g. “change the color of the sky to bright blue”).

  • It can also follow a range of different prompts to generate short or long captions and answer questions about an image.

  • The presentation sounds intriguing, but assessing the model's actual performance is challenging until it becomes available to the public.

• Stability AI introduced Stable Doodle, an innovative sketch-to-image tool that transforms basic drawings into dynamic images, unlocking boundless imaging potential for professionals and enthusiasts across various fields.

  • Stable Doodle is available to try for free on the Clipdrop by Stability AI website.

  • Under the hood, Stable Doodle seamlessly integrates Stability AI's state-of-the-art image-generating technology, Stable Diffusion XL, with the formidable T2I-Adapter. This powerful condition control solution enables precise command over AI image generation, granting users meticulous control and customization options.

• Bard from Google received several exciting updates this month, including text-to-speech functionality in 40 languages, response modification, conversation sharing, and more. Two updates caught my attention in particular:

  • Bard is finally available in the EU.

  • It has been integrated with Google Lens.

    • Now, users can upload images along with text in their conversations with this product.

    • For instance, you can upload an image and ask for a description of its style. This feature can serve as a viable alternative to Midjourney's /describe command, which occasionally produces peculiar outcomes.

📌 AI Art Tutorial: Testing AI Art Generators

Check out this video to see how the same prompt works in various AI image generators, such as Stable Diffusion, Midjourney, Adobe Firefly, Bing, DALL-E, and more. Many of these generators are accessible for free, allowing you to test them yourself for your specific use cases.

🎨 Featured Artist: Dehiscence

Dehiscence is a Toronto-based, multi-media artist whose main subject matter is the human form. Her artistic process stems from a desire to transcend the prejudices and stereotypes that are so commonly associated with our external physical traits and to highlight social factors which impact human rights, inclusion, and equality.

She started experimenting with AI in 2018 and has been using AI infusion since 2020 as part of her process to edit and manipulate her collages or digital sketches. Dehiscence entered the NFT space in March of 2021. Her artworks have been shown extensively throughout North America and Europe.

🖼 AI-Assisted Artwork of the Week

🤓 A Comprehensive Midjourney Guide

Share Kiki and Mozart

If you enjoy this newsletter and know someone who might also appreciate it, please feel free to share it with them. Let's spread the word about AI art and introduce more people to this fascinating field!