Whats the Prompt? Dreamstudio + Stable Diffusion Public Release - Follow-up Post about AI.


It has actually been 2 crazy weeks since I started this experiment and this post is about what I learned.

On a positive note, dreamstudio which is the website editor based on stable diffusion, is now in open beta publicly, so you may try it on your own with 200 free credits.

And you can download and or get your own Stable diffusion running as google colab notebook or on your own PC if you know how. You can join the discord to get help and answers to your questions: https://discord.gg/stablediffusion

Here is my previous post with more discussion material and my opinion from around 10 days ago.

The 2 weeks beta-testing on the stable-diffusion discord has been an awesome experience, it was really an enrichment, so much creativity and inspiration - I have not had such a feeling since I started studying art back in 2003!

The weights of stable diffusion have been released and it can be used in google colab or locally or via the dreamstudio website. The unedited output I have posted over the past weeks can now be done or used by anyone. I really like the community spirit and want to give something back, so if you like to create simple concepts in the style like above, here it is: 

greg manchess portrait painting of ...your character... as overwatch character, asymmetrical, profile image, organic painting, dynamic pose, sunny day, loose brush strokes, hard edges, conceptart, art by gil elvgren and ruan jia

CFG-Scale around 8.0 sometimes 7 sometimes 9 and steps work well with 50 up to 90 and only up to 576x676px.

I believe it's a great benchmark style and I learned a lot through it and use altered versions now, but the base should work just fine. 

If you use it, a mention is appreciated but not necessary.

Why I slow down on posting

If you want to follow my output of these experiments, I will keep up a folder on my artstation / website fantasio.work that is named AI-Art / Experiments. For now that is the only place I will keep posting besides deviantArt and sometimes IG.

I will not stop posting art in general, but the reason why I want to take a break with posting  AI-Art is due to many issues:

Since the release of the weights and dreamstudio, anyone can do what I did in the beta. I need to find ways to incorporate this technology in my existing work and new styles to set myself apart.

As far as the public perception goes, there are heated discussions on youtube, deviantArt, instagram and reddit. I also have discussions with friends and other artists and it seems many concerns and problems stem from misunderstandings and maybe mis-interpretation.

Creating AI Art and publishing AI Art are two very different things it seems.

Here is a great example of mis-interpretation:


He is bashing literally every art director out there and if you ever worked for a big company, you know there are art directors who can "prompt" an artist greatly and others not, so IT IS an art-form, just a different one, like cooking. Knowing a lot about lighting, art history and materials. I'd go as far and say you need to be an artist for quite some time to be a great Art Director aka "Prompter" ;)

From my perspective, I can say that I have a certain knowledge and passion for artists and art history that gives me a big advantage in this game. And in general my stance is very similar to what Olivio Sarikas speaks about in his short 7 minute video here, highly recommended to watch:


My main problem with publishing is that I can not be 100% certain that everyone looking at the art knows that it is generated or if that change the perception and value of it?

I mean, IT IS a matter of perception; If I would tell someone that I did not paint all my work but Photoshop did, (that is not possible as of today but you get the idea) would that devalue my work?

But since I put my hands on it and countless hours does that give it value?
Or the 20 years experience with digital art?
Is this discussion even relevant considering that a chair with fat (aka ready-mades) is considered art?

If in the future Photoshop is A.I. driven and can paint on its own with just peaking a prompt, do I have to tell everyone that I did paint with my hands back in the day? Or does no one care anyways? We don't know.

What I know is, that these are important and sometimes philosophical questions that have to be answered all individually - which takes time and effort.


Social Media and Statistic insights

Personally, I will refrain from posting on IG and FB because I believe most people don't really understand it or are wondering and I don't want to add to the confusion. There are also other reasons.

From the social media statistics, I had much hope on Instagram but not so much on Artstation or deviantArt but the opposite is true.


Just a little statistic-insight; I started with that experiment posting the "ready-mades" on Instagram with over 600% of reach on the first day of posting and 200% interacting accounts opposed to the month prior. That went down to just 100% after a week. New followers went up about 0.5% at the peak which is where organic reach (without influencers) has dropped

The system is really broken. I remember early last year when running an ad campaign for 25 € reach on one single post was around 25.000 up to 65.000. Imagining that it will become even worse when everyone will post AI-generated art very soon and a lack of human curation, that will not make it easier to be visible there AT ALL.

This also answers a big question to me - if Instagram is still worth the time and maybe money for ads in 2022? My takeaway is to keep using it for posting not so often but with quality content for existing watchers. Actually like before the experiment, the result of this experiment is reason enough to focus on other platforms.

Also interesting to note is that publishing from Instagram to twitter, pushes only 1 image and somehow that resulted in no reach at all. Twitter may just give exposure for organic reach. Learned that the hard way. 

Somehow the same for tumblr posted through IG, some traction maybe but not so much, not sure if tumblr is still alive anyways.

deviantArt led to 5000-7000 views on individual images opposed to my own portrait work that just had around 700-900 but I suppose that is because of the hashtags which are trending right now. I will be careful about that and only upload an AI piece here and there in between my regular work, at least dA isn't broken when it comes to hashtag reach.

Artstation led to a steady 1000 - 1500 views per day (starting from an average of 500), which is not bad at all - and that is for all pieces in the gallery. 

So in my opinion it will take a shift in platforms more towards curated and specialized art platforms. I hate to say it, but maybe NFT-sites are better suited or be pushed there anyways.


Technical restrictions of A.I. at the moment

The thing I, as also many other artists and enthusiasts have hoped for, was higher resolution.
But working and testing Stable-Diffusion in beta and words from the developers answering questions the last couple of days, made it very clear, that the model is just capable of working perfect in around 512 x 512 px.

That is the reason I have not much tested for landscapes. Portraits turned out perfectly well in square but produced broken images about 60% of the time on aspect ratios like 2:3, 3:4 or more.

Also the rendering time goes up exponentially when requesting images with many steps and resolution of around 512 x 1024, which makes it take even longer on your own hardware.

As it is now, I will try to get it running as a google Colab Notebook and train it with my own data when the weights are being released and will keep my eye open for this kind of development, but that might take at least 2-5 years. Maybe earlier when upscaling software like Topaz Gigapixel uses the technology to get better at upscaling too.

A great thing is the init-image option and I guess the community is looking much forward to an inpainting feature soon!


Mistrust

One thing that I really have a problem with, are artists pretending their work was done in Photoshop, when it is clearly obvious it was 80 % stable diffusion and they did only some touch-ups in PS. 

No name-calling here but it is to be expected that this will happen much more often and is also a reason I want to limit reach to my little own experiments, because those artists give the tools a really bad name. But they don't think that far ahead. Time will show them and a backlash eventually.


Quo Vadis?

Where will I take this? I'm not so sure. For the time being, I will keep experimenting with these tools until they get better and until people get used to it. 

For the portrait works, I might definitely take time to push my own boundaries and see where that takes me and how to set myself apart. I don't have a problem creating new art-styles but it takes time to create a coherent series. Maybe AI can help me to create the graphic novel I have in mind for Luna & Nova.

I will definitely keep up works on my Stilllife-series and the Ancient Kaiju Project, everything else is just cherry on top for now.

Let me know what your experience looks like on social media or how you go about it on a daily basis.






Tags: , , , , , , , , , , , ,

Written by

Oliver aka Fantasio is a creative blogger who likes to share his insights about art, marketing and social media. Follow Fantasio on twitter or facebook

0 comments:

Post a Comment

Comments are moderated

 

Popular Blog Posts

Recent Blog Posts

...
Copyright © The Art of Fantasio | Terms + Conditions