Ugh, I hate University group assignments/projects. I'm currently doing a Master's course and have a group assignment due tomorrow, but all the answers/work that I've received so far from my fellow team members have ZERO in-text citations. And these folks are doing a post-grad course. Fuck, whatever, it's the last semester and I just wanna get this course over with. Online classes suck ass, too.
Text to Art via AI interpretation is pretty damned impressive. It's also impressive how far they've come in the last few years.
The text in the tweet title are the words given to the AI to generate the image. 'Tis mad.
Been playing around with it for a while now (trying out different parameters and keywords) and I finally got something decent.
dragons in the sky blue lightning | artstation | unreal engine
Parameters for the above image: (also used sflckr as the image source. Which I assume means Flickr.)
Did a second attempt with more parameters:
beautiful dragons in the sky with blue lightning in the background | artstation | unreal engine
Parameters for the above image:
An AI is able to do something I could never. I can't even draw proper shapes for the life of me. Bloody amazing. We're moving ever so slowly to a future where Skynet will exist.
For what I've understood so far, the number of iterations you ask the program to perform is important. Higher the iteration count, better the final image. Also, sticking to a low res image is a good idea if you don't have google collab pro. I used 500x500 & 600x600. Kept getting GPU out of memory errors anything above those values.
I recently upgraded my internet plan. Going from 100/40mb to 250/25mb. The upload speed is kind of a downgrade, but it's a sacrifice I'm willing to make in order for Holotools not to crash and burn during mega-collabs.