I listened to another long audio chat from David Holz last week. He was thinking auto-generating 3d models from text is going to be next and not that big of a leap from generating 2d images.
ooh, I said this’d go into 3D right when it first hit. That one will be cool.
but then this is a learning algorithm that learn on art it gets feed it [so it does better that more fresh art its feed], what if it eventually feed loop it with its own generated image. - eventually one of its own generated image will get feed back to it.
Also something I thought day 1, after seeing Dall-E 2. I’m expecting the output may get narrowed by the influx of same-style images, but since there’s a lot of generative tools and they all have their own styles, the influx will be varied enough to prevent total tunneling.
What are people’s thoughts on artificial intelligence being used to create artwork
Ah, I had many thoughts on this and it took a week to sort out how I felt.
Officially speaking, it’s a good development that’ll give non-artists the ability to visualize thoughts and give artists a more powerful workflow, help them grow, get used for inspiration and draw-overs, etc.
But unnoficially, I understand the sense of being cheated. There’s some disillusionment for the insecure artist who sees this all and wonders “well, what’s the point if this AI is better than me”. And the fact it’s trained on skilled artist/photographer output makes it a “biting the hand that feeds” situation. Capitalism takes the path of least resistance; the breadth of the art industry will shrink to some degree here, and/or pay less. I think that’s reality.
Some forms of recourse for the dreading artist:
The best way out is to draw for a reason that isn’t peer approval. Drawing for your career / as a gift / for meaning / for fun / being social are all valid. And if you find you’re stuck seeking peer approval, think about this: there’s too much “better” art out there to even view in your lifetime, already. All the AI does is make this obvious.
Have a medium, style or subject matter that’s hard to one-button reproduce, e.g. comic book lineart. Traditional artwork is safe because prints cost money and people will value traditional medium originals > AI prints any day. Some mediums, like wood art or painting on objects, can’t get reproduced at all
One of the thing i keep thinking about that need to be settle is copyright or copyleftness of this images.
Who owns the image generated by this resources? are they even copyrightable?
Are they copyleft? i.e. if someone generate an image and posted it - can i then can pick that up and and use it as inspiration with minimal attribution [who do I attribute this to?]
If you feed the words - [what do you own, the words used or the image?]
This is for sure a live issue. Recently the consensus seems to have become that, in the USA at least, AI creations are not copyrightable. I think this case might be the source of that:
But that case may not be applicable to these prompt-based tools, since it hinges on the requirement of human involvement in creating copyrightable works. It doesn’t seem reasonable to me to say that a human directing the output of a computer tool could never have meaningful involvement in the creative process.
This is one where i think - complicates. Which is the copyrightable one - the prompt or the image generated or both. I feel that the same prompt can produce different image depending on when it was queried since its learning. How much plagiarism/ copying can be said if two prompts produce similar result?
i feel like people are dancing around this issue lol.
In a way - the copyrightability of the work is I think one thing that help artists and squarely put this tech as a tool because you will need a person to make changes/ produce artwork for the copyright.
If the AI generated works cant copyright hold true - the debate will shift how much change from the generated image will now then constitute enough change to be consider “human made” to be copyrighted.
One of my main dilemma now - is seeing really nice generated images by other and getting some “ah i want to take that and make something out of it”. should i consider every image i see generated as public domain for now, free to take? I kinda feel that they should be - public domain.
how different is inputting a prompt? to inputting numbers to generate fractals in a fractal genarator?
so many question.
omg pardon me - my mind is running full steam ahead on this technicality.
It’s interesting to me as technically, it’s a derived work that should list all copyrighted sources under fair use, at least per the typical U.S. copyright policy. I’d say so long as it’s not almost identical to any one image, it’s good enough to be fair use.
I’m guessing “it can’t be copyrighted” is a temporary ruling to avoid lawsuits in the meantime, which might not align with the final ruling after this concern has been given some time.
As far as I can tell, based on the article above, that ruling was in response to someone asking if works created by an algorithm designed to require minimal human input to create art can be copyrighted. This same person previously tried to get a patent assigned to an AI for some invention created by an algorithm, so you can maybe see what this is about.
It was not a ruling about humans using AI tools to create art, nor a ruling about whether e.g. giving an AI tool text input is sufficiently “creative” to justify copyright protection. I haven’t been able to find any cases anywhere on that, so I guess, despite all the reporting, we’re still firmly in “wait and see” territory here.
It doesn’t matter what the “copyright” law says. If the tech is here, it will hurt your job opportunities eventually, it’s just a matter of time. If it can lower the labor costs for the investors in the entertainment industry, it will happen.
My biggest concern is that this AI art creation will eventually take away the fun of doing digital painting.
It looks like one person already made a plugin for Krita that is using the “Stable Diffusion” AI for helping with art. Here is the plugin. I think they are still working on the instructions a bit.
For me it matters, yeah it stays here - but if im going to use this as a tool then i need to know the copyright grounds. Copyright and Attribution, if i generate something or I pick something from someone who do the generating thing, what would be the licensing fees [of the images not the AI]? Is this like stock image, or pose packs, background packs kinda thing. Are they public domain - do I pay the company who owns the AI - if I pay those company licensing fee , will the artist whose work was use to feed the AI get some kind of royalty fee?
Things like that? there is a whole legalese landmine there.
I think in the future this issue will be sorted out. For example, what I can imagine is that corporations could hire a lot of artists to create enough source images, then use AI creation internally for their customers, (I assume those AI generated images with or without paint fixed can also be used as source images, just keep multiplying).
As time goes by, the need of artists will be diminished…and the pay rate will be much lower because it’s much easier to create beautiful paintings.
- eventually one of its own generated image will get feed back to it.
I kinda feel that they should be - public domain.
there is a whole legalese landmine there.