The Remnants and a mini VQGAN+CLIP guide

in Alien Art Hive3 years ago (edited)

Hello everyone. Hope you are all doing fine and thank you so much for your overwhelming support. This is a post where I describe how I got started with AI art and how anyone else interested can do so too.

It all started with this post https://peakd.com/hive-158694/@jotakrevs/minds-dungeon-ai-artwork-engspa#@lavista/re-jotakrevs-qygij2 around a month ago. I commented on the post showing my interest in the process and @jotakrevs so kindly replied and directed me to YouTube and wished me luck. Off I went, searching for tutorials on what VQGAN+CLIP was about and how one could go about generating such amazing art using AI.


I am a firm believer in the use of technology and science to solve all problems and make our lives better. I often quote this famous quote by Arthur C Clarke :

Any sufficiently advanced technology is indistinguishable from magic.

And I have been fascinated by the recent advances in machine learning and use of AI in the medical field where a sufficiently trained model can predict cardiovascular risk by just looking at pictures of the retina of humans. This is just an example related to my field of work, Ophthalmology. Similar things are being done to identify whether excised specimen tissues are benign or malignant by training models. There are fallacies of course. No new technology is perfect but this is magic for me!

Forgive my extensive deviation from the topic at hand.

Imagine my joy when I discovered art could be made with the help of AI. This is because I can hardly draw; I had to draw a mammal when I was in school and I of course drew a dog. I thought that was that. A dog is a mammal and in my mind that was the correct answer. I did not understand art then. The dog I drew was emaciated with sharp corners at all angles. It was far away from the word 'cute'. My artistically inclined sibling and mom had a good laugh looking at it. That is my background with art.

It does not mean that I did not appreciate art. I always marvelled at the thought processes going through the minds of great artists when they produced what they produced. The pandemic has been a blessing in that regard, at least for me, where I tried spending my free time trying to create stuff.

Long story short, I tried my hand at 2D and 3D art and have been able to 'create' some stuff but there is a lot of room for improvement. Hence, I tried my hand at AI art.


These are the YouTube videos that I saw initially and they can help you get started and a little more. I suggest that you watch the third video after you have tried the VQGAN+CLIP notebook and played around a little on your own and have seen the first 2 videos. Also, I suggest that you try the jotakrev's Enhanced original VQGAN+CLIP notebook initially, which is first in his list of notebooks, that he provides in all his recent posts. Do take a look at all his work, they are awesome.


There is no programming knowledge required to do all this, if you are worried about that. If you want to modify the notebooks, yes, it is required.

After I started posting stuff that I generated, I got great support here in the Alien Art Hive community. They really are very supportive and encouraging. It was @castleberry 's idea that we sync our findings together and it has already worked wonders.

Just as I was wrapping up this post, he has posted his first post involving AI art, his methods and experiments and there are videos too to check out, a comprehensive post. As always, he has been very generous in his mentions and comments on the work of his friends :). Here's his post.

Check out @dbddv01 's post here :https://peakd.com/hive-158694/@dbddv01/my-creative-process-with-ai-generative-tools-for-drafting-the-broozel-city-guide and also check out his other amazing posts. There is a very important post regarding ownership of AI generated art, which is necessary reading for everyone getting into this.

Next, after you have played around with the original VQGAN+CLIP notebook, check this post out by @kaliyuga for more fun: https://peakd.com/hive-177949/@kaliyuga/kaliyugas-library-of-ai-google-colab-notebooks.

This is the twitter feed of the author of the original VQGAN+CLIP and CLIP-guided diffusion notebooks: https://twitter.com/RiversHaveWings.

Do check @eve66 's posts too for some more great AI art and other wonderful collages.

There is something called as ArtBreeder too, it was introduced to me by @emptyintentions. It is much easier to use than VQGAN+CLIP, if that is what you want.

I try to use images generated on artbreeder as initial images in VQGAN+CLIP and find that I get more consistent results. After using the original VQGAN CLIP notebook, I suggest that you try the MSE weighted notebook by jotakrevs in the link mentioned above. It reduces the variations consistently and helps in getting more coherent images. You can use the REAL ESRGAN notebook in his list of notebooks to enlarge images with AI. I find the website https://bigjpg.com/ very helpful too, in that regard.

Also, if you want a 16:9 ratio for your images, you can use upto 736 x 414 pixels size without getting any out of memory error messages even on free Google Colab.

I think @kitzune and @chocolatescorpi might be interested in the links in this post.

This is not a typical Alien Art post, forgive me; hope people wanting to get started with AI art find it helpful. I end this post by attaching some of my AI generated images which did not fit in other posts and I still found them beautiful. Thanks a lot and see you in the next one!


EDIT : The video links posted here are not made by me, they are videos posted by others, I have linked them here as they are good starting points while using VQGAN+CLIP.

bay2_out.png

cada1.png

dacosm2.png

das1.png

daveg1.png

Sort:  
 3 years ago (edited) 

I think your background in Ophthalmology is a really interesting angle to approach this process artistically. I bet you know words for the prompt most of us would need a degree or two in order to even understand! hahaha I was digging your deviation up there very much! Thanks for such a comprehensive post as well. I think the more open and free we share our experiences and methods the more comfortable folks become in exploring this artistic adventure of ours! Awesome outcomes with some of your tinkering! Keep up the great work!
i forgot to mention that that Clarke quote was awesome. I had that same idea when i was young and used to share that sentiment. I never knew it had been articulated so well already before!

I actually haven’t used any medical terms or even any technical words in my prompts I think; thanks for the idea :)😁.
Thank you for reading, am glad that you like the outcomes.
Yes, any new subject is difficult for newcomers mostly because there is no portal of entry ;), all most newcomers see are huge walls making it difficult to get into the subject. I hope we can make it easier for everybody to try out this stuff.
Our posts were so similar in spirit and the mentions are almost the same! If this was an exam and if I was a teacher, I would have said we have copied or discussed the topic before writing the answer 😂
I have been a fan of Clarke from a long time and I feel his writing was way ahead of his time. Like you said in your post that everybody has some art in them and later some move away in adult life, I believe it is the same with science. Every child is born inquisitive, every action they perform is a physics experiment and as we become adults, many of us move away from the magic that is science. Anyone can always come back :)
Thank you @castleberry for reading and appreciating my posts, it means a lot :)

That is a very interesting observation about children and science... because when i think about it, Art is Science and Science is Art. I don't think i would have come to that conclusion without your statement there.

I have always loved eyes and the science of how we see. It's extremely fascinating. I won't get us going about that... but it surely could supplement the text prompts. =)

Glad i didn't have to grade our papers... because i would have assumed copycats! hahaha I forgot one metion of eve66. She has some fantastic art and started a bit before us in the process.

Science fiction is definitely one of my favorite literature genres. Asimov and Clarke were well ahead of their (and perhaps our) time.
Take Care!

Art is Science and Science is Art

I now have another awesome quote to use :).
True, regarding Asimov and Clarke.
Thanks, you take care too...

Wow that a sufficiently trained model can predict cardiovascular risk by just looking at pictures of the retina of humans I would really like to hear about that!

And thank you so much for such a detailed and informative post!

Without people like you sharing info like this, then dummies like me would never have a chance to try ourselves!

So I don't know if I told you, that I created an acc with Artbreeder and downloaded blender too and have been watching heaps of tutorials for both, but have a doco captioning job to finish (hopefully tomorrow) before I can really start playing with them, but you keep motivating me with all of you info, hints, tips and tricks, so thank you once again!!!

AND bookmarked

When we look at the retina after dilating a patient's pupils, we look for changes in the blood vessels in the retina in addition to other findings. For example, arterial narrowing is usually seen in hypertensive retinopathy, basically uncontrolled hypertensive patients. Machine learning is probably able to correlate the magnitude of narrowing of blood vessels to different grades of cardiovascular risk. Maybe changes too early to detect by humans or changes too minute to grade by humans can be done easily by training a machine learning model.
EDIT : @dbddv01 has kindly shared a link regarding use of AI in ophthalmology, you might find this interesting!

Thank you so much for your kind words. I am a beginner myself. You did tell me that you created an account with Artbreeder and that you were interested in learning Blender. Yeah, so hope you finish your work soon and can have fun later. Thanks for reading my post and for the bookmark :)

Huh. WOW. That is fantastic!

I will definitely checkout @dbddv01 link as I am one of the millions of women around the world- especially Australia, where heart problems do not get considered, tested, diagnosed- even AFTER heart attacks.

There has been so much new evidence over the past few years (and even decades) with cardiologists- only some and not many in Australia....that are finally acknowledging that women have heart attacks which has nothing to do with a build up of plaque in their arteries, so I wonder if this kind of ai learning could detect a person with Printemetals Angina (first found in 1959 by Dr Printzmetal) AKA Variant Angina AKA Unstable Angina AKA CAVS- Coronary Artery Vaso Spasms....Imagine if it could...that would be wonderful for soooo many people around the world, then they wouldn't have to go through a decade of being ignored, mistreated and disrespected by arrogant cardiologists and doctors like I have...

Yeah so creativity here we come....😃

Get regular ECGs done at least if not anything else.
Looking forward to your art now:)

Hi @lavista, will check the videos and try the 16:9 size, good tips ! and yeah, i won't expand on it but definetely AI recognition process in Healthcare is a big leap forward. Tons of improvement everyday, especially in your domain of work.

Yeah, it is just amazing what technology can help us do in healthcare. I had written a post about how I thought 5G can help us, quite some time ago.

Thank you for reading my post :) your work and explanation of all your processes is awesome! and thanks so much for the link :)

I was about to comment in your previous post as these artworks are really beautiful. But now I found that got mentioned here hehe Anyway, I really loved what you did there!

And thanks a lot for the mention... this is awesome, who would expect that give you some advice that day, will become this growing group of AI artist. Really awesome!

I'm amazed with how much detailed are your artworks, even when you used the diffusion, diffusion is awesome but usually blurry and not very detailed.

Great artworks as always, thank you and grab this !PIZZA!

Haha, I am one of very few people who is not overly critical of his own work 😂. And here, it is mostly AI work, which I find really beautiful :)
Thank you very much for your appreciation, support and yeah, without that initial post of yours and directions, I might not even have tried AI art. Thank you for everything :)

I am one of very few people who is not overly critical of his own work.

Good for you, never change haha

You are welcome (:

PIZZA!
PIZZA Holders sent $PIZZA tips in this post's comments:
@jotakrevs(2/10) tipped @lavista (x1)
Learn more at https://hive.pizza.

Thanks for mentioning me! I’ll be definitely trying all these soon. Your posts have been inspiring so I’m looking forward to create some crazy stuff 😆 Let’s see how it goes. Also your story is really interesting, I’m glad you were able to discover art using this method, since your field of work is totally different from this 🙂

Hi, thanks for reading :) I am glad that my posts have inspired you to try some stuff! I am happy too to have tried AI art and it definitely has 'prompted' me to write more! A decade back, I was definitely not an arts person...