In this post i wanted to share with you some of the A.I. experiments i have been making. Once again, without the help of some friends, i wouldn't have the videos to share with you. So, i will be adding that friend as a beneficiary for this post.
I will try and keep my words a bit more brief than usual since there is so much to see. I don't normally save the static images since it's way more fun to see the images slightly change from one iteration to another in the videos. I think it's easier to see what is changing as the subject of the prompt morphs into what it is we are both trying to accomplish.
Throughout the post, i will share some static images so you can see some of the processes without having to necessarily watch the videos!
I titled this post the way i did... not because Mona Lisa is anywhere in here, but like her notorious smile (or lack thereof) has been a topic of focus for hundreds of years. I thought the portraits cam out exquisitely gorgeous and these smiles are quite beautiful.
They tend to start out with an old classic art feel and then morph into something else. I actually prefer some of the beginning iterations better than the end result... but i couldn't resist letting the cycles run.
I was under the impression, like all the other notebooks that the seed parameter set to -1 always randomly generates the seed. Apparently, i was wrong. I was starting to become astonished at the consistency of the outcomes.
I was almost beginning to think the A.I. had a crush on me by providing such beautiful results! That was also not the case but wishful thinking perhaps. I assure you... there are no nipples this time! The A.I. cut me some slack finally.
You can check out the Nvidia StyleGan3 with CLIP guidance Notebook here at this link. You just basically run the cells.
When you get to the Run Setting cell you delete the url target embed links and delete the target = clip_model.embed_text and add your own text in between the quotation marks. You can also put in a number for your seed and play around.
I had a great time running these experiments since it's much faster than the usual notebook and the results are undeniably stunning!
Each change from one iteration to the next is usually very subtle. All of these static images in this post are from runs i have been doing while i am getting this post put together. Good thing i remembered my seed! hahahaha
I wasn't always happy with the final outcomes but who can complain really when it runs fast and a vast majority of the time produces excellent results?!
I could usually tell after a few passes if it was going to be worth the wait or not. I ended many runs early to tweek the settings a bit and save my GPU usage.
I figured after doing so many with red hair... why not try some with blue hair?!
I have always been partial to blue and red hair, not necessarily on the same head at the same time.
I did try a couple of runs to include blue hair with red streaks and vice versa... but i don't think the A.I. was ready for that level of creativity!
I do think that the blue hair runs made the A.I. get creative and take some liberties with the eyes. Anything that adds interest to the artworks is fine with me.
I changed the seed for the static images to try and get a bit more diversity for your viewing pleasures.
The videos were already created based on previous runs over the last week or two, when i still thought -1 seed was random. I did kind of enjoy seeing the consistency in the runs since it's so rare working with this medium.
The static images turned out really nice. It was interesting to see a new face. I was very fortunate to get almost all the portrait subjects to smile digitally.
If you notice in the videos... it seemed like they were hiding. Peeking around the edges... or starting in the middle and then trying to work their way off the screen.
I really found the face of this static subject to be quite beautiful. It's hard to think or imagine that this person does not exist. I wish i could jump inside the mind of the A.I. as it does it's iterations and see where and how it's gathering data and creating this person.
The weirdest thing is that these ladies seemed very familiar to me. All of them for different reasons. I say all because even though i didn't change the seed there were about 2 or 3 other generated subjects. However, 99% of the time they didn't pan out well at all. You can see one such different subject in the Green Woman video below.
So, to conclude this post i am going to just put a run of images from the final experiment functioning in the background as i put this together. I hope you have enjoyed the massive amount of imagery and scrolling. Now, just scroll a little bit more. =)
Well, it worked out perfectly. I don't know if you noticed or not but the smile just kept getting bigger and bigger in that run. It's hard not to enjoy seeing the beauty in that process.
Thank you very much for your time and finger muscles scrolling through this post and enjoying some A.I. generative art. If you read... give your eyes some credit too! Y'all take care until i see you next.
