Remixing your art with AI

2456721

Comments

  • generalgameplayinggeneralgameplaying Posts: 517
    edited October 2022

    Write Idea said:

    Not to mention some also make your artwork communal property.

    Using AI to assist, of course, is not special in terms of "copyright or not", assuming functions like assisting with something like denoising an image or helping with texturing a model. Using entire generated images, or large portions of those, even if they are made up on base of your ingenious texts, however, might become a different thing.

    It's quite certain, that images generated by AI will not be possible to copyright, because it would be fairly easy to prevent anyone from being allowed to do anything ever again, just by flooding the world with generated images  (short and simplified version), if you allow AI to produce copyrighted material. Maybe some day a "strong AI" in terms of intelligence similar to human intelligence, will be a different case. 

    Of course it will be hard or even impossible to distinguish, to what extent you used AI-generated content for the general case, and you may end up on the safe side with altering and contextualizing such generated images. Maybe it's possible to backtrack parts of the images, though, and to detect "behavior" of the AI-system, so one wouldn't necessarily be on the safe side, with claming not to have used AI.

    I just want to point out, that law is not fully decided on, and in the end, it might turn out to be different in different parts of the world.

    Post edited by generalgameplaying on
  • WendyLuvsCatzWendyLuvsCatz Posts: 38,198

    video 

    can get endless characters from one render

  • cgidesigncgidesign Posts: 442
    edited October 2022

    Fun fact:

    I tried a local instance of stable diffusion and got an image with "Shutterstock" watermark in it. That did not come from the prompt but seems to part of the AI training base model. Question now is:

    The AI companies say: "you can legally sell the AI images" but on the other hand they use copyrighted material in their training data which you can not sell without permission. In some cases you are even not allowed to use images from stock sites at all without permission (e.g. a subscription plan for personal use or so). But if I now publish in the DAZ gallery or sell somewhere, won't I this be potentially already a copyright theft issue?

    As far as I know Getty and Shutterstock already removed AI images from their sites because of that.

    But, it is really fun using those tools. Below images are from the Midjourney beta discord server (no remixing here though).

    The attached images are of higher resolution.

    elve_girl_in_forrest_water_pond_translucent.png
    1664 x 1664 - 3M
    elve_girl_in_forrest_water_pond_translucent_2.png
    1664 x 1664 - 3M
    Post edited by cgidesign on
  • ed3Ded3D Posts: 2,177

    ~  And  an interesting subject nonetheless   ~

  • Worlds_EdgeWorlds_Edge Posts: 2,152

    Write Idea said:

    Some of the end results I've seen look great.  I have always wanted to draw and/or paint well, but I just don't have the talent.  That is why I use DAZ to help bring my ideas to life.  I would use these, but there are some major drawbacks.  These AI companies are charging monthly subscriptions or micro-transactions for one render.  Not to mention some also make your artwork communal property.  If a company offered a stand alone product for a flat price, I would buy it.  I'm not interested in spending $30 a month on a hobby.

    sign up for artbreeder. They have now added stable diffusion. It is free. 

  • Worlds_EdgeWorlds_Edge Posts: 2,152

    Write Idea said:

    Some of the end results I've seen look great.  I have always wanted to draw and/or paint well, but I just don't have the talent.  That is why I use DAZ to help bring my ideas to life.  I would use these, but there are some major drawbacks.  These AI companies are charging monthly subscriptions or micro-transactions for one render.  Not to mention some also make your artwork communal property.  If a company offered a stand alone product for a flat price, I would buy it.  I'm not interested in spending $30 a month on a hobby.

    sign up for artbreeder. They have now added stable diffusion. It is free. 

  • I jus' thought'f/f... stablediffusion and the likes can alter one image, but can/what it with two?

    Remember science to cover up for basic counting abilities of some animals, literally taking thousands of years, where would THIS lead?

  • McGyverMcGyver Posts: 7,050

    Well, one thing that's helpful about AI art generators is that it's gonna make for more interesting Christmas cards this year...

  • MelissaGTMelissaGT Posts: 2,611

    cgidesign said:

    Fun fact:

    I tried a local instance of stable diffusion and got an image with "Shutterstock" watermark in it. That did not come from the prompt but seems to part of the AI training base model. Question now is:

    The AI companies say: "you can legally sell the AI images" but on the other hand they use copyrighted material in their training data which you can not sell without permission. In some cases you are even not allowed to use images from stock sites at all without permission (e.g. a subscription plan for personal use or so). But if I now publish in the DAZ gallery or sell somewhere, won't I this be potentially already a copyright theft issue?

    As far as I know Getty and Shutterstock already removed AI images from their sites because of that.

    One of the first images I created with my Stable Diffusion free credits developed with what was quite obviously a checkered watermark layered in..meaning the AI pulled a stock photo from a site like Shutterstock as part of it's source material. Bad, bad juju. That means none of our own artwork is safe as they're likely out there being pulled in as source images as well. *sigh* 

  • Zombieklaus :).

  • bytescapesbytescapes Posts: 1,837

    cgidesign said:

    I tried a local instance of stable diffusion and got an image with "Shutterstock" watermark in it. That did not come from the prompt but seems to part of the AI training base model. Question now is:

    The AI companies say: "you can legally sell the AI images" but on the other hand they use copyrighted material in their training data ...

    This is definitely the case.

    Apparently a lot of current artists are unhappy with the situation, because users will write a prompt that says "An ogre eating a pineapple in the style of Artist McArtistperson" and the app will then generate images based heavily on sampled works from Artist McArtistperson, reproducing their distinct style. I don't see an issue with that if you're generating images for your own entertainment, but apparently some people are starting to sell the generated images online and even advertising them as being "by Artist McArtistperson". I can understand why the artists are upset.

    ... won't ... this be potentially already a copyright theft issue?

    I can well imagine that we'll start seeing some lawsuits.

    The AI companies may come under pressure to remove certain content from their training sets. This is going to be tough for them: if Artist McArtistperson complains, the AI company will not only have to pull their work from the training set, but they'll probably also have to rebuild the model. Based on what I know of the technology, I believe that this will likely require them to recompile the entire model, which takes insane amounts of computing power and is thus very expensive.

    A cheaper alternative would be to leave the models as they are but simply add "Artist McArtistperson" to a list of keywords that are filtered out of the prompts. So if you ask for "A bowl of angry bananas by Artist McArtistperson", you'll just get images of "A bowl of angry bananas".

    I wonder if we'll see a throwback to the early days of rap music, when musicians started putting "All samples cleared" on their albums?

    But, it is really fun using those tools. Below images are from the Midjourney beta discord server (no remixing here though).

    Those are lovely. And the human figures are impressive. Both Midjourney and Stable Diffusion, in my experience, have a lot of problems with human figures ("How many legs do people have? Three, isn't it? Does three sound right to you?") But those worked very well.

  • CrystalwizardCrystalwizard Posts: 102
    edited December 2022

    Write Idea said:

    Some of the end results I've seen look great.  I have always wanted to draw and/or paint well, but I just don't have the talent.  That is why I use DAZ to help bring my ideas to life.  I would use these, but there are some major drawbacks.  These AI companies are charging monthly subscriptions or micro-transactions for one render.  Not to mention some also make your artwork communal property.  If a company offered a stand alone product for a flat price, I would buy it.  I'm not interested in spending $30 a month on a hobby.

    https://www.mage.space/ is free.

    As are the Stable diffusion demos on huggingface:

    Stable Diffusion 2.1 https://huggingface.co/spaces/stabilityai/stable-diffusion ;

    Stable Diffusion 1.5 https://huggingface.co/spaces/stabilityai/stable-diffusion-1

    and I'll be more than happy to help you learn to prompt craft if you want. There's a learning curve to writing a prompt so you get images you want as opposed to 'what is that supposed to be" images. here's a prompt to start with for 1.5

    elaborate ink illustration made of glassy liquid

    put that in, hit generate, then maybe add a subject such as sphere to get

    elaborate ink illustration of a sphere made of glassy liquid

    Post edited by Crystalwizard on
  • kyoto kidkyoto kid Posts: 41,034

    ...we finally have the long awiated "Make Art" button. 

  • I have a few that came out pretty cool 

    I'd love an outfit to be based on some of these too 

     

     

      

     

     

  • KitsumoKitsumo Posts: 1,215
    edited December 2022

    Write Idea said:

    Some of the end results I've seen look great.  I have always wanted to draw and/or paint well, but I just don't have the talent.  That is why I use DAZ to help bring my ideas to life.  I would use these, but there are some major drawbacks.  These AI companies are charging monthly subscriptions or micro-transactions for one render.  Not to mention some also make your artwork communal property.  If a company offered a stand alone product for a flat price, I would buy it.  I'm not interested in spending $30 a month on a hobby.

    I know I may be late to the party, but have you tried a local install of Stable Diffusion? You use your own hardware, and have a lot more features to use. If you're rendering with a GPU in Iray, then you can most likely run SD. Personally, I use Automatic1111, but there are others.

    Automatic1111 offers most of the usual offline features, inpainting, outpainting, upscaling, you can use models trained by other people (anime, particular artist, Breaking Bad characters or whatever), and it includes Textual Inversion which lets you train your own models (but not as sophisticated as Dreambooth). All pictures are stored on your hard drive.

    It can also create tileable textures. You can just ask it to create a texture of a parquet wooden floor or damaged sheet metal with rust stains. It may take a few tries, but there's a batch feature so you can click generate and play a game or something while you wait.

    Edit: I'm on Linux and it was relatively pain-free to install. I understand Windows is a bit more work and I haven't tried it, but from everything I've heard, it's still worth it.

    Post edited by Kitsumo on
  • Video

    Merry ai Christmas

    original video used

     

    Carrara 2010

  • When I buy something on this site to use, I am getting a license that says I have permission to use it. These AI art engines have learned based on stolen art that was not licensed, credited, or reimbursed for that purpose. If these models were scrapped and new models were built off of data sets that were licensed or people opted in to, it would be a different situation entirely. But trying to say that using DAZ products is the same is wrong.
  • SnowSultanSnowSultan Posts: 3,595
    edited December 2022

    I didn't say it was the same thing, but I always find it funny when people here complain about AI art not being 'real' art when we can buy every element to create a scene in this store, all created by someone else. We are the last ones who should be yelling about things like 'the death of art' and 'no effort is required'. It's often harder to get really good art out of an AI than it is to assemble some quality content from DAZ into a good looking scene.

    Also, AI art isn't really stolen, as artistic styles cannot be copyrighted and the AI does not literally piece together recognizable bits of existing art. The terms and service notices on websites that no one ever reads often allow for data to be sold for information purposes, which is apparently how a lot of the images that were scraped in the initial training set were legally obtained. But I'm not here to argue that, I'm just disagreeing with that earlier statement that those who think AIs have a place in the art world are absolute scum.

    Post edited by SnowSultan on
  • wolf359wolf359 Posts: 3,827

    kyoto kid said:

    ...we finally have the long awiated "Make Art" button. 

    Yes quite literally!!

  • SnowSultanSnowSultan Posts: 3,595
    edited December 2022

    I don't know if I'd go THAT far. Being able to draw and paint are still amazing talents and I do not wish those people to be put out of work because of AI. I also hope that people who use AI will specify when they do and that they will make some effort to personalize the end result with their own touches.

    It's the constant lies about AI that bug me. It's not stolen, it's not all ugly, and it's not easy to get results that can rival those of a trained artist. I basically have to do what a photobasher does, as well as a lot of actual manual drawing to get results that I'm satisfied with, and it's partly why I don't like being lumped in with other AI artists who crank out a dozen images of fantasy girls with huge breasts for their Patreon pages - but if people are willing to pay for that and they enjoy it, why should they do without? How many completely new careers were both created and made obsolete by the growth of the internet?

    I get both sides of the argument, but I see AI as a tool that should be used to help artists and enable those with less artistic talents to be able to bring their creations to life as well. For people who claim to be all about artistic freedom and against censorship, there sure are a lot of them who want to decide what is art and what isn't.

    Post edited by frank0314 on
  • mwokeemwokee Posts: 1,275

    Write Idea said:

    Some of the end results I've seen look great.  I have always wanted to draw and/or paint well, but I just don't have the talent.  That is why I use DAZ to help bring my ideas to life.  I would use these, but there are some major drawbacks.  These AI companies are charging monthly subscriptions or micro-transactions for one render.  Not to mention some also make your artwork communal property.  If a company offered a stand alone product for a flat price, I would buy it.  I'm not interested in spending $30 a month on a hobby.

    It's all fairly new and will probably evolve. I looked at one application, it was $30/month for CPU time and to have your images kept private and avoid being being communal property. But to justify the cost, I would have to keep cranking out images and I don't have time beyond my current projects. AI doesn't give me a final product, I need to do post-processing to accomplish what I want. That adds even more time but the good is it makes it nearly impossible for others to duplicate. The costs of the current services will most likely go down, I just hope it's soon. But man, to render an image in minutes compared to the time it takes to do a traditional 3D render..! The tests I did were better than Daz, at least for certain types of artwork. If I get a lull I may sign up for a month and see what I can do, investing $30 in my business isn't going to hurt.

  • I've seen some pretty unexpected fallout from the AI thing.

    There is an artist that I follow on Face Book that paints. Now they spend their time going back and forth with viewers who believe the painted images are from an AI engine..

     

     

  • HeraHera Posts: 1,957

    Where do you find the software to do this?
    I've played with NighCafe and some other similar tools, but I pretty soon got tired of dysmorphic faces and hands and how everything manmade (like vehicles and buildings and such) looked malformed. And I've gotten nowhere to get the result looking anywhere similar to the original picture. 

    So if someone can tip me about good engines out there, I'd be a lot pleased!

  • wolf359wolf359 Posts: 3,827

    Hera said:

    Where do you find the software to do this?
    I've played with NighCafe and some other similar tools, but I pretty soon got tired of dysmorphic faces and hands and how everything manmade (like vehicles and buildings and such) looked malformed. And I've gotten nowhere to get the result looking anywhere similar to the original picture. 

    So if someone can tip me about good engines out there, I'd be a lot pleased!

     

    https://www.creativebloq.com/news/ai-art-generator-comparison

  • generalgameplayinggeneralgameplaying Posts: 517
    edited December 2022

    SnowSultan said:

    I see AI as a tool

    Right. For most. On the other hand "ai" is a buzzword as well and suggests more than there is (machine learning algorithms for instance).

    The problem with the training data remains, and may be taken on by lawmakers. The outputs can also be in question, though it'll be pretty easy to move your creations into grey zones, i would assume. 

    You can't defend "ai", which is another important thing to note. We are dealing with specific implementations of concepts of machine learning, and it makes all the difference, what they actually do and what they are trained from. So i wouldn't say the stealing point is without any base, not taking too much of a side, but imagine the dominant cloud-service providing the text-to-image functions and more, training on artists works without their consent, effectively reducing the market share of said artists by a lot. Also note, that the alternative for an artist would be not to publish, with this taken to the extreme. This is just one thinkable scenario, there may be others, including widespread free tool availability. In my opinion lawmakers really need to create something meaningful here, or you'll just have the next medieval prince on top of some cloud. Independently from that, there will be tools. That's certain, and there will be very good tools for all sorts of things, concerning 3D, animation, games, movies. Question is, at what cost (and for whom).

    Post edited by generalgameplaying on
  • wolf359wolf359 Posts: 3,827
    edited December 2022

    In my opinion lawmakers really need to create something meaningful here, 

     

     

    …which “Law makers” will have the global jurisdiction to implement & enforce their “meaningful “ legislation to cherry pick training data
    already in the hands of private entities?
     

     

     

      

     

     

    Post edited by wolf359 on
  • generalgameplayinggeneralgameplaying Posts: 517
    edited December 2022

    wolf359 said:

    …which “Law makers” will have the global jurisdiction to implement & enforce their “meaningful “ legislation to cherry pick training data
    already in the hands of private entities?

     

    EU can. You don't walk along, you don't sell there. Of course USA can. (...). China can try ;). Social media get blocked everywhere for anything already, it won't necessarily be different with art.

    Yes someone can build the ai, and it becomes a question of distinguishability. Though there may be rules imposed on social networks, concerning scraping, so at least from some point on, your ai will be short of data. It's not certain yet, what will happen.

    It's basic law to enforce "explicit consent only"... [difficult to enforce, though, yet... embargos on art... why not?]

     

    (And yes: you could argue, that if we don't try slave holdery v2 now, we won't know if it actually could work out for us. Doesn't work for maintaining a civilization, conceptually, though.)

    (Not to appear too cocky here, but ... artists might have a choice, at least this opens the business model for the checkbox on allow-use-for-ai-training. Once an algorithm does something like weigh in choices by ratings, it's no difference to checking for consent anymore. Meaning: This is a societal choice with societal impact.)

    Post edited by generalgameplaying on
  • As I said in anothr thread, if potential commissioners of art consider AI to be high risk (because of potential copyrightn liability in their jurisdiction, or jurisdictions in which they wish to be active, or because they will not be able to protect their own AI-generated images from uses that may damage their brand) then they will not accept AI-generated art. No global legislation would be required, merely local legislation in desirable markets.

  • ServantServant Posts: 758

    SnowSultan said:


    It's the constant lies about AI that bug me. It's not stolen, it's not all ugly, and it's not easy to get results that can rival those of a trained artist.

    It actually IS stolen art, photos, graphics, etc. Midjourney confirmed they did not acquire the rights needed, because the A.I. (in its current form) requires MILLIONS of datasets for the machine learning to produce anything beyond an amorphous blob. So they did the shortcut of simply NOT GIVING A DAMN about getting permissions and/or compensating the authors they took the data from.

    No photo description available.

  • Servant said:

    SnowSultan said:


    It's the constant lies about AI that bug me. It's not stolen, it's not all ugly, and it's not easy to get results that can rival those of a trained artist.

    It actually IS stolen art, photos, graphics, etc. Midjourney confirmed they did not acquire the rights needed, because the A.I. (in its current form) requires MILLIONS of datasets for the machine learning to produce anything beyond an amorphous blob. So they did the shortcut of simply NOT GIVING A DAMN about getting permissions and/or compensating the authors they took the data from.

    No photo description available.

    While that is an appalling attitude, and I hope they get mangled in court for it, we should remember that AI is more than any particular engine.

    If AI-generated imagery can be protected by copyright and/or if the original copyright persists through generation, then I can certainly imagine big IP companies making at least some use of the technology, trained on thir own assets, if only for one-shot and incidental imagery.

Sign In or Register to comment.