Today I had an idea of what I wanted to achieve in Midjourney AI art creator. I remember myself being in Shibuya in Tokyo. I was alone but was overwhelmed, excited, and energized by the amazing street views at night.
I wanted to get that feeling: a shy Japanese woman with a strong presence and a very mysterious and electrifying personality. I wanted a woman to really stand out like time stops and she is the only thing that matters in the scene.
By the way, the Kawaii keyword at the beginning of the description tends to make the characters look much more painting-like.
I remember when I was celebrating Halloween in Osaka on Dotonburi street. I saw a beautiful woman, stunning and my jaw dropped by her beauty and presence.
It wasn’t easy getting Midjourney AI to generate the art I want. Because the scene is very complex, it’s not like I am going to change in retouching things in Photoshop. I would probably ruin the integrity of the unique look of the scene or/and the character. So I Was counting on the Deep Learning artificial intelligence to do the work in a way that’ll be satisfied with.
One of the main issues was again the face. It turned out to be horrific in many renders. I added “cute face” to try to stabilize the results and hopefully, the pathing will lead to trained images where the face looks more symmetric, bright, and clear. It helped to some degree. Some faces still looked like they were taken from a horror movie. No way I can fix that face with all the lighting going on—I did make a few face fixing after all LOL. I even did Fill Content-Aware in Photoshop to remove and reposition the lips.
I knew that for this one, I’ll spend quite a lot of time, but I liked the overall artistic results I knew I am going to stick with it, until my AI art looks great.
Sometimes, when you add too much lengthy description, the AI just goes nuts. So I try my best to keep it short and descriptive–or at least to see if its solves some issues. If you are to vogue in your description of what you want in the AI art, you usually get “bad results”.
The overall image should be fashionable, powerful, cinematic, and more realistic yet partially painting but a bit.
The modeling keyword should make the girl move like a model or post like one. It also affected the facial expression and clothing the girl is wearing. This added more dynamic and the overall look appeared more seductive and dynamic. And sometimes it leads to a complete octopus-like mess of a pose.
I also used an object selection tool to separate the model from the background and apply stronger vibrancy without affecting the model itself, just for the background. Also separated the model and apply higher brightness. Lots of work on this one.
Also, sometimes the girl pose is amazing, but you get a totally ruined face or even sometimes, no face at all. So I must use Photoshop in order to mix and match because I might not get that type of pose anytime soon.
I also used the word Lightning. I like strokes of lightning appearing in random ways, it makes the whole scene more manga-like and energized.