[Image Generation AI - Part 1] Midjourney for Creators: How Far Can We Express Our Ideals?

Introduction

Hello! I'm Yuri from OpenFashion’s PR team.

To everyone reading this blog, thank you for enjoying the OpenFashion site. At OpenFashion, I work tirelessly day and night to publicize not only our services but also the charm of generative AI to a variety of people, not just those in the fashion industry.

Now I strongly feel the potential of "Generative AI is amazing!" However, before the AI fashion platform "OpenFashion," where this blog is published, was released, my knowledge level was at "Generative AI...??", "P, prompt…?? Sounds tasty..."

Yes, I was an “AI novice”. Lol.
 

How is Image Generation AI Actually Used?
Let’s Interview the Users

Even an “AI novice” like me has become able to express through AI easily since I started using the image generation AI "Midjourney". I think many of you visiting this site are already familiar with it. Midjourney is a service that generates images via Discord when you input prompts (instructions or questions).

Just by typing a prompt, artwork to my taste is generated in about 30 seconds...!Initially, when I started using this image generation AI "Midjourney", I was enjoying the “accidental encounters”, wondering "What can be generated with this prompt?...", but as I kept using it, my desires grew.

"I want to generate what I'm picturing in my head..."
I believe some readers of this blog might feel the same way.

Sorry for the long introduction, but today, I want to ask people at OpenFashion who are utilizing Midjourney,

・Can they clearly imagine what to generate beforehand?
・What do they imagine when generating?
・What do they refer to when generating?

From the perspective of a generative AI beginner like me, let’s hear from them!
By learning the tricks of generation from OpenFashion members who create numerous splendid pieces with generative AI, surely both you and I can express upgraded works with image generation AI. Now, let's dive in!

Introducing the Participants

The roundtable participants are MASAYO, RIE, and SAKI being interviewed, with YURI as the interviewer, making four in total.

👩 MASAYO (Interviewee)
・In charge of production in the CG team
・Started CG after joining OpenFashion from the apparel industry
・Favorite prompt words are "hyper realistic", "toy core"
・Favorite generated image and prompt:
Three fashion models wearing unique details outfits, virtual fashion, near future, curvilinear details, sophisticated, simple Background, hyper realistic.

👩 RIE (Interviewee)
・Belongs to the CG team and is in charge of PMO (Project Management Office)
・Hasn’t been involved in design since her student days, though she comes from an apparel background
・Wants to create a picture book for her child with images made by Midjourney
・Favorite prompt words are "unique", "sculpture", "big silhouette"
・Favorite generated image and prompt: She likes images that are "cute, but a bit interesting".Full body composition High fashion editorial photography, studio photography, black puffy coat collection, in the style of sculpture, high tonal range, vibrant color choices, modular design, clean-lined, Sweden

👩 SAKI (Interviewee)
・A web industry designer, recently handling screen design (UI) and experience design (UX)
・Has no special knowledge of apparel, a complete novice
・Favorite prompt words are "abstract", "geometric", "canvas"
・Favorite generated image and prompt:It's a photorealistic interpretation of melon cream soda, vanilla ice cream, and cherries that are vividly rendered, but still retain the muted pastel geometry. When viewed through a camera lens, the bubbles appear hazy, capturing depth of field against a fragmented pastel blue sky and adjusting focus to create a photographic bokeh effect. (Note: These are three different prompts).

👩‍🦰 YURI (Interviewer)
・PR at OpenFashion
・Originally was in an IT company but is a "AI beginner" still learning about AI
・Strives to explain things in a way that even those not familiar with the topic can become interested
・Favorite prompt words are "cotton candy", "twinkle"
・Favorite generated image and prompt:

Full body photo, three fashion models in dresses made of metallic material.One person's motifs are pastel pink, ribbons, candies, and rubies; the second person's motifs are pastel blue, clouds, waves, and sapphires; the third person's motifs are pastel yellow, soap bubbles, and yellow diamonds

"Let's Add More Words to Get Closer to What We Want"

YURI: Thank you for joining today!

So to jump right in, do all of you have a specific image in mind that you expect to get when you input certain prompts when generating images?

MASAYO: It’s somewhat vague, but I do have my own logic, like bringing themes or motifs that are high in priority and I don’t want to deviate from, to the beginning of the prompt.
So I can halfway predict, “If I put in this instruction, something like this will probably be generated”, but for the other half, I can’t really tell what will come out.

SAKI: I vaguely have an idea of “maybe something like this will be generated,” but I hardly have a firm image of “I want to bring out something like this!” from the initial stage. I just input the keywords, look at the generated piece, and think, “let’s add this word to get closer to what I want.”

For instance, this generated image of a dress. I made it when I started using Midjourney because I like dresses.Initially, when I input “onepiece,” nothing came out. But when I input “dress (a short skirt length),” a silhouette to my liking appeared. In this way, I had decided on the base form of the clothing, but other than that, I had not decided on anything else besides thinking “since it’s April, let’s add some cherry blossoms.” I hadn’t imagined how it would be generated.

YURI: Interesting! By the way, how did you come up with these boots?

SAKI: There’s a piece of clothing that serves as the basis for this boot design. I wanted to make boots that incorporated elements of that clothing, so I extracted the features of the clothing using Midjourney’s “describe*” feature and inserted the words that came out into the prompt for the boot image.

*describe: A function of Midjourney that reads an image and generates its prompt.
 

Everyday “Random Thoughts” Turned into Artworks

YURI: How about you, MASAYO? Your works are quite unique.
What process do you go through to create them?

MASAYO: When I made this piece, the situation with COVID-19 had calmed down quite a bit and I was tired from going to the office for the first time in a while, haha.

I didn’t want to bump into the many people at the station, and my feet hurt from walking so much, so I felt like floating in the air. From there, I thought, “I want to go to the office floating on air with shoes that can float,” and this artwork was born. There are also prompts like “No stress” included.

SAKI: It’s amazing to reflect everyday random thoughts into images through generative AI!

YURI: Indeed. And there are various processes before generating, like using describe to extract words, or first creating the type of clothing and then adding motifs. Above all, it’s fascinating that interesting pieces are born just by directing prompts from what was thought of in everyday life.

Thank you, everyone!

To Be Continued in the Next Part

Since this blog is the first half, we’ve heard what the participating members think about while generating. In the second half, we plan to delve a little deeper into their thoughts.

・Do they feel they’ve mastered using Midjourney?
・How can they generate the images they want to create?

Stay tuned for the next part!