Artificial Intelligence Art (MidJourney, Dall-E etc etc discussion)

[Ryan_Knight]

I just found out about this yesterday...

https://en.wikipedia.org/wiki/Artifi...telligence_art

... the existence of typing in some words to get some high quality images is (as an aspiring professional comic artist) mind-blowing to say the least ...

I know the age-old argument: 'technology has always changed art' but... a program that can create art that takes seconds is probably pretty game changing... I'm truly surprised how fast this tech has arrived... I never thought something like this would exist so early in my life... AI Art can create something in a mere moment that even the most adept photo-basher takes hours to make...

Some people are even making incredible looking comics...



I mean sure, for the trained eye, you can kind of see that the art seems to be a mixing or a montage of photoshop-like filters and all that jazz but... I feel a bit torn by this discovery... Even though a huge side of me is upset that someone would have the shameless desire to create a program like this, part of me is saying 'wow I can probably use this to make some cool looking backgrounds'...

Would love to read anyone's thoughts on this...

[Ryan_Knight]

Here's another image from a comic a user made

[Ryan_Knight]

Some people have even printed their comics

[Ryan_Knight]

it's crazy ...

[battlewraith]

Yeah it's really advancing quickly. I did a beta for midjourney and it was interesting but not all that impressive. I did a second beta a few months later and the results were significantly better. Also there is now a community of people on youtube that are doing tutorials on how to more effectively use prompts and how to use different applications that do different things in conjunction with each other to create better images. The most interesting thing to me at the moment is the use of image weighting. So some of the programs can start off of an existing image rather than creating something from scratch. So an artist might be able to do a rough sketch or simplified version of a picture and then use the AI to add more detail and elements to the composition. Imagine wanting to do some urban landscape but not having the patience to place cars, pedestrians, signage, etc in the picture. Maybe you could place some basic features and have the program fill in the blanks. That would be immensely useful.

[Ryan_Knight]

yeah it's insane how fast it is progressing... check out this video on reddit ... I still can't stop trying to figure out how it's doing this

https://www.reddit.com/r/StableDiffusion/comments/wyduk1/show_rstablediffusion_integrating_sd_in_photoshop/

[Ryan_Knight]

Here's a longer one on StableDiffusion with more info on YouTube

https://www.youtube.com/watch?v=xbxe-x6wvRw

[D.C.]

I tried the Img2Img Stable Diffusion using parameters that I saw someone else get good results with using one of my sketches and I gotta say, this was not worth the hour long wait.

What have they done to my boy?

I was browsing the Stable Diffusion subreddit though, and some of the results are scarily good. (I hate how much I like some of these Invincibles.) Not seeing this replacing comics anytime soon (doesn't seem to be able to recreate the same character from different angles with different facial expressions with any sort of consistency based on the examples I've seen so far) but for concept art, it feels like it could potentially put people out of work.

[ChrisH]

I was browsing the Stable Diffusion subreddit though, and some of the results are scarily good. (I hate how much I like some of these Invincibles.) Not seeing this replacing comics anytime soon (doesn't seem to be able to recreate the same character from different angles with different facial expressions with any sort of consistency based on the examples I've seen so far) but for concept art, it feels like it could potentially put people out of work.

I've been playing around with it too. I don't think it will stop the real creative types, since it can only imitate it's only going to copy existing styles and not invent a style. Also, there's little to no panel to panel storytelling in the comics I've seen. Being able to draw the same character from different angles is an issue that will probably be solved, but it's not good at dynamic shots of people running and jumping either. I think it will allow people who have no artistic ability to create, and I can't see that as a bad thing. I'd be interested in using AI art as reference images instead of copying existing artists, it certainly is good at composition most times. But you have to know how to draw since it really doesn't keep track of basic anatomy and how many hands, fingers, and arms a human has X-D

[battlewraith]

What I'd actually like to use something like Stable Diffusion for is to fill in background detail. So I would do an image and keep the characters and things that are actually important to me, while using the program to fill in the blanks and render out the stuff that I either don't like to draw or forget to put in anyway. And then composite the various iterations together in Photoshop. But I haven't gotten around to experimenting with it yet.

[Ryan_Knight]

Someone already copyrighted their midjourney created comic

Artist receives first known US copyright registration for latent diffusion AI art

[EddieChingLives]

https://youtu.be/tjj6KsPSHZc

[Ryan_Knight]

Hi there again guys. (Merry Christmas for those who celebrate Christmas)

A lot has changed since we last chatted about the topic, for one I found that the AI is powered by copyrighted art. I went from ‘wow, it’s a great toy’ to ‘man, robots are evil’ in a few months.

Let me explain:

I found a lot of human artwork used on the LAION Database (the training database for AI generated Art used by Stable Diffusion) this is what is used to power AI Art… the AI learns from our art. I am basically an unknown artist and I found around 70 of my images, and I'm still looking. If you choose to opt out any of your copyrighted images (there are many personal pictures on there as well) and/or are you're an artist and want to opt out go to haveibeentrained.com and search for your images (I’ve found that uploading the images works better) and right click (on desktop) and choose to opt out. This website is created by two artists from Germany who wants the people behind AI generated Art to make more ethical models, where they create AI art with datasets that only include consensual data. When the images fed into the programs are not consensual, I think this is when the issue becomes a very problematic and ethical nightmare (among many other things) because, it’s kind of like stealing an artist’s DNA without their permission. Here’s a print screen of my art that I found, I’m sure some of your art is on there too:

[Ryan_Knight]

I made this comic early this morning

[Ryan_Knight]

Stop the Nonconsensual Use of Artists' Copyrighted Artwork/Images for AI Generated Art!

Hello once again all my fellow penciljackers, it’s time to Stop the Nonconsensual Use of Artists' Copyrighted Artwork/Images for AI Generated Art!.

Please sign the petition if you have a minute, you can log in with your google or facebook account.

Change.org



Here’s the petition’s description:

Popular AI Generated Art programs, models and/or applications such as MidJourney, StableDiffusion, DALL-E, Lensa and many others have used unethically scraped copyrighted data from the internet to create digital art. Once again these Big Tech Companies are benefiting commercially from the data we upload online. Big companies SHOULD be held accountable for violating our rights as artists and most importantly humans. It is understood that without this copyrighted data, the programs, models and/or applications could not have made the beautiful artwork it is creating. Among the datasets used for this machine learning, are the LAION-5B Datasets which contain over 5 Billion Images, many (if not most) that are Copyrighted material. Many artists are in agreement that this Data was taken without their knowledge or consent. As of this writing, the burden is on the owners of the copyrighted material to opt-out through a very meticulous process on the haveibeentrained.com website, this has been the only recourse provided and many artists do not have the time to find all of their images. Also, many artists believe that even if they opt-out of future datasets, the damage is already done and that this will not change anything since the aforementioned programs, models and/or applications have already been trained on the copyrighted data. Humans should be protected from technologies that potentially have an adverse effect on humankind. Please help us STOP the Nonconsensual Use of Artists' Copyrighted Artwork/Images for AI Generated Art AND help us RESET the data that has already been used so that the aforementioned programs, models, applications (or anything else that uses this data) be re-trained with data that does not violate copyrights.