Things have been wild in the AI generative community over the last few months. All the models are developing at breakneck speeds and we are achieving some really amazing generated content in seconds. I am still in awe as I witness the changes happening in the image generation subsection of AI. It’s also just so much fun to just throw prompts in and parse through and refine results. You can see the clear difference between the generated work I am able to prompt today, versus just a few months ago.
Here’s some of the work I’ve been able to generate, with relatively simple prompts. If you’re curious about the prompts, just check out my midjourney profile. It’s public!
You can see that facial features and consistency between variations has greatly improved.
A really neat new feature is that you can zoom out a photo and the surroundings will be rendered! Rendering the 2nd image above with more environment, zoomed out at 2x, gives us a really neat depth of field effect that takes inspiration from the colors and shapes of the original!
There’s also a new style to use, named Niji, which uses models trained on Japanese animation — coupled with the ability to change the aspect ratio to your liking. This has led me to make really nice wallpapers for my computer and phone!
I honestly can’t stop playing with the color limits and the portraiture features. It’s hard to believe this variety inspired by the same prompt is possible. This tech is in its toddler days, so I am both stoked and a little apprehensive to see what happens in the next 2-5 years.
I am a huge fan of sailor moon, and have been since I was 10. You know what’s awesome? Sailor moon in Indian attire!! I can only dream this up, but couldn’t execute it myself before AI without probably decade+ of art study and practice. Now I can try it in a few seconds and honestly be totally in love with it. How powerful!!
Of course, we can really take it to a different style and try fun manga cover style and zoom out a bit.
Full Body Images geared towards more realistic models are also spectacular!! Swapping color schemes on the subjects gives it a real creative twist.
Culturaly specific images are also getting better. Previously, trying to get a sari to generate well was a real task. Now, its pretty good on the first try.
Textures and landscapes that are illustrated have made for some great backdrops for my devices. It’s very easy to produce an entire theme with image inspiration.
I’m able to produce some urban street scenes too — loads of details are now rendering discretely instead of melding into one another.
Image generation is getting better and better — and it’s taken the cost and speed of first pass ideation to almost nothing. I’ve really enjoyed the opportunity to use this tech!