My Experience with Midjourney AI
Trying to design a Steampunk Zelda from the Twilight Princess gameFeb 22, 2023
DISCLAIMER: I understand that these image AI tools are being trained with artists' imagery, without their consent, and that is a whole legal and ethical can of worms I am not looking to open in this blog post. I think it is wrong and that we need legislation to protect artists' work. The purpose of this article is to simply share my personal experience with using generative AI.
As someone who has a passion for cosplay and costume design, I was intrigued when I first heard about Midjourney AI, a generative AI technology that produces images based on keywords and themes. A lot of cosplayers love creating unique, anachronistic, or cross universe, versions of our favorite characters, and I wanted to see if AI could help me do just that by designing a steampunk version of Zelda, from the Twilight Princess game. Below, I want to share my experience with Midjourney AI, and my thoughts on the AI technology.
Background on AI Technology
Artificial intelligence, or AI, refers to the ability of machines to perform tasks that would normally require human intelligence, such as understanding natural language, recognizing patterns, and making decisions. AI has been around for decades, but has become more prevalent in recent years, as advances in computer hardware and software have made it possible to build more sophisticated AI systems.
Recently, tools like ChatGPT, Midjourney AI, DALL-E, and many others have become main stream and are already revolutionizing the way we work, live, and create. These tools are here to stay, so I think we should all figure out how to live with them and what their strengths and limitations are.
Personal Experience with Midjourney AI
I first came across AI being used in costume design by Firefly Path. On her Instagram, she was sharing how she was trying to get Midjourney AI to create a dress design in her style that she could then create in real life. I thought this blending between the digital and physical world was so cool that I wanted to try it out for myself! (Read more about her project here)
For this experiment, I wanted to use Midjourney AI to generate costume inspirations for a steampunk Princess Zelda. The process was simple: I entered a few keywords related to the theme, and the AI generated four ideas. I could then either upscale one, use one as a prompt to get four more images, have regenerate the images using the same prompt, or ask for a new prompt all together. I started out simple and direct, and what the AI produced was also pretty direct (though I wouldn't call the design simple):
Not bad, but not quite what I was looking for. I then proceeded to go down a Midjourney rabbit hole and spent a few hours fine tuning the prompts, upscaling images, to see how far I could push the software and what kind of variety I could get out of it. Below is small sample of some of what I was able to achieve:
I noticed that since I wasn't specific about Princess Zelda from Twilight Princess, it was sometimes generating an image with lots of blue and gold: very Breath of the Wild inspired. Midjourney AI allows you to use an external image as a prompt, so I was curious what would happen if I fed it one of the original concept arts from the game:
It definitely captured the painted style of the original art, which I liked, but it also ended up producing a lot of images that were too similar to each other in terms of design and style, which I guess would be expected:
A Closer Look
All of these images were very different and could be used to make very different cosplays. Do I want something that looks like a Medieval knight? Or a Victorian lady? Or steamship captain? Or something more traditionally steampunk? I then started looking at some of favorite images in more details and trying to understand how I could bring one of these to life. And this is where I understood exactly why AI won't be replacing real artists anytime soon.
None, not one of the generated images, was actually feasible to create as is. They just didn't make sense. For example, look at these um, swords:
If I'm out adventuring, I want to have a nice utility belt with me with pouches, and potions, and other useful tools, but this doesn't make sense:
Or what about the lights on this skirt. They definitely look cool, but why are they there in the first place? What is their purpose? They seem to big and heavy to just be decorative gems so they should be some kind of battery/power source, but then why would they be on flimsy skirt?
And below is a sample of some other weird issues that I saw:
What also really got to me is the fact that the triforce symbol was missing from pretty much every single image. It is an iconic symbol in all the Zelda games, and is always featured on Zelda's outfit, but here it was nowhere to be found.
Basically, every single image was missing a story and context. None of the created characters had usable weapons, armors, tools, or looked like they were full, complete creations. If a human artist was tasked with creating such a character, I'm confident they would have understood how to make a character that made sense by creating a story, a world around them, and making them instantly recognizable.
Final Thoughts
This is where I understood that as of the writing of this article, AI Image Generating tools have their limitations. For example, Midjourney AI seemed to struggle with understanding the context of the fantasy theme, and some of the costume ideas it generated were simply unrealistic. Additionally, while the AI was good at generating a large volume of ideas quickly, it couldn't replace the human intuition and creativity that goes into designing a truly unique and meaningful costume.
However, I found that Midjoureny was incredible at creating images that can be used for inspiration. I believe that AI technology has the potential to help designers overcome creative blocks, generating new ideas, and assisting in the design process. In fact there are several images that I would like to combine together, and fine tune to potentially create an actually feasible cosplay. The tool was able create some very cool looking, and incredibly different designs for a character. I was definitely inspired and felt like I had some great ideas I could use as a starting point to create something really awesome.
2022 was the year that AI technology went mainstream and became easily accessible to everyone, not just engineers. It's controversial, it's fast, it's terrifying, but whether we like it or not, it's here to stay and only going to get smarter from here on out. It's a brave new world, and we have to learn to live in it.