Stunning Hydrogel Art by Artificial Intelligence

In this newsletter edition, I’d like to share some image generations that I have obtained from Midjourney.

Midjourney is a text-to-image generation software housed in the Discord application. We can create images by telling the Midjourney Bot a sequence of descriptors. Descriptors are words or phrases that describe your desired image. There are a ton of modifiers you can add into the descriptors and I have not fully explored them all.

Staying within the theme of this newsletter, here are some hydrogel-based images from Midjourney!

Nature-cover Inspired Image Generation

I wanted to see whether I could create a journal cover by feeding the Bot a sample journal cover. This is called inspiration, a method where you send the Bot an image url as the prompt. This is the nature cover I supplied it:

And here are the descriptors that I added behind the supplied image url.

url hydrogels, microstructure, atoms and bonds, chemistry, futuristic, technological

Midjourney took the url and descriptors into account and generated four image previews. It then gave me the option of upscaling or making variations of the following sample grid.

Midjourney added a journal name to each image, an artifact most likely inspired by the supplied image url. It did not seem to understand “microstructure”, so I told it to give me atoms and bonds. This resulted in a network-like image that is often seen in hydrogels.

I am slightly disappointed by the result as it is closer to a virus model than a hydrogel network. It is obvious that the training data used for Midjourney contained by conventional virus images that look like a spiky ball.

Original Promptcraft Images

Promptcrafting is a new word that describes how we think of new descriptors for the AI to produce better images.

Looking at the previous inspired image, I felt that there was too much influence from the words in the image (nature title, research paper captions). Thus, I wrote new descriptors from scratch in my next attempt:

hydrogel, microstructure, atoms, chemical bonds, molecules, surface properties, tribology, adhesion, friction, stiffness, amonton's law, veiny, poster, scientific

This is the grid preview that Midjourney provided.

I liked image #1 the most because it looked realistic. Images #2 and #3 were acceptable too, but seemed flat, less dynamic than #1. Finally, #4 started to look like comic book art so I did not like it. Desiring a finalized product, I asked the Bot to upscale #1.

Honestly, this is a really beautiful final image. I hope to share this on Twitter someday, but thought that I would let our newsletter subscribers have the first look!

My final extension was to make this image even higher resolution. This idea came because of YouTube videos from MKBHD and Linus Tech Tips, who film high resolution videos for the platform. I also saw other users in Midjourney Discord channel adding the descriptor “8k” in their prompts.

To add this new descriptor, I used the Variation function on the previous image. Here is the result:

The image looks similar to before. I think the soft water-like reflections have become sharper and start to look like a honeycomb. The bonds between droplets also start to look more like roads. This might be due to the predominance of architectural and cityscape photographs used in the Midjourney training dataset. Perhaps I need to add more modifiers in front of “8k” to avoid that influence.

Have You Tried Midjourney?

I’ve shared some beautiful images in this newsletter edition. The software I used, Midjourney, is by no means new in the technology world. It was first released in July 2022.

Some of our readers might have tried it already. I’m curious to know what you feel about the results. Also, did anyone try using it for science illustrations like I did in this newsletter? Hit ‘reply’ and let me know!