… an astronaut riding a horse in space
… a mountains lake view crystal clear sky night stars
Always curious always learning I started tinkering with AI Art generators. I have a good knowledge of how Artificial Intelligence and Neural Networks work and explored their potential years back for speech and image recognition. But little did I know how advanced certain networks have become over time to solve for specific areas like generative art. Indeed it is a creative explosion and one can easily be pulled into the rabbit hole and never come back.
Check it out
I tried the generative art on one of my standard profile pictures with quite interesting results: https://hummelimages.com/galleries/imagine_yourself/
How it works
The network of an AI art generator is informed with millions of images and concepts. Think of it as a person that you can talk to. Everything you say or show to it will trigger an association, combine concepts and give back what it thinks is a fair result. So playing around with it and thinking what you could come up with as an input prompt, you quickly realize:
It’s all about the right question
Every input prompts are composed of several elements. The keywords that you give to describe the concepts and visuals that you want to trigger. Can be text but also a link to an already existing image. You can add shortcuts to access certain settings of aspect ratio or quality.
- No output is the same. It is like you talk to a human being that also develops over time. It learns with every input.
- Every network has distinct flaws. E.g. while using midjourney you see how it pivots to a standard visual of a women (the midjourney woman). Even if you use a male portrait picture you sometimes get a femal looking result. That nicely shows the bias that the network brings in.
- The possibilities are endless it seems as you can type in whatever you want. However after a certain point in time you realize that with all the possibilities still you are limited in the boudaries of the framework or its interpretation of your prompts.
It’ll be interesting to see similar techniques be applied to other creative and design areas. Why not compose music in this manner? Why not code an application with just text prompts? Why limit the concept to the creation of 2D images? Why not apply it to create complex functional 3D machinery (I know this might require huge calculation capacities.
Let me know what you think of AI generated Art.