I mean, the hostility is entirely understandable. The current form of generative art is meant to replace artists. It is part of what is currently devastating peoples livelihoods, although I think some companies and clients are already learning that it currently leads to lower overall quality, due to how much harder it is to implement changes based on feedback. It lowers the overall quality bar, although it does have the potential to raise the floor a little. The larger models that are causing this hype are quite literally trained on the work of unwilling artists.

It is the most disrespectful and clearly ethically wrong basis to build it on, and it really begs the question of whether the ends justify the means. Beyond that, art is just not an area where we need AI. It largely hurts artists, is super energy demanding so it actively hurts the environment for no real benefit.

The energy would be so much better used solving actual problems, so more people could spend time doing things they enjoy. If some people enjoy AI generation, then that’s fine but I think it shouldn’t replace a passionate, skill-based workforce.

That’s a cool visualisation of what kind of visual input you can feed into the process with ControlNet.

And it really makes it clear that what AI images is good for if communicating a general idea. I think comparing AI generated or Assisted images or videos to photography is probably the closest analogous medium we have, but I think AI images are stort if in-between that and more classical art. You have more control over the more technical aspects of the image, as you can alter those things with big strokes, but you’ve given up too much control to really infused it with artistic intent. Even when photography, where you are generally limited by reality, you can better infused artistic intent into the picture, because you carefully examine what makes that object of the picture unique. Even if you try to direct AI models, it limit their scope they will always add whether the most average expression of what they’re adding, because that what it looks for in the training; the commonalities/averages of whatever it was trained on.

Even ControlNet is just a way to claw back a little more control over the process. I wouldn’t actually call the examples I’ve seen of ControlNet to be examples of fine control. I’m struggling to find a way to clearly communicate it, but it’s like the difference between 3D art that is trying to look like 2D, and actual 2D. There’s always something lost in the translation.

Most artistic disciplines are their own language, and I just don’t think we have a way to communicate that language without actually doing the art, and art requires artistic intent, which I don’t think is possible with the current AI tools. Maybe it will be at some point, but artistic intent and control over the process are so interconnected that the balance becomes very difficult.

That sounds like a good tool for you then. I do art as my day to day, and there are definitely aspects of my work that I would love to have a bit of AI injection to help speed the process up, but that is much more as a tool directly integrated into the softwares I already use, like a beefed up content aware transform tool that allows me to move parts of a finished image around just a little bit, and having the AI fill in the small gaps that creates.

I see so many small ways to make the art process less painful or introduce more non-destructive editing tools, if only the AI was built into the software and actively training on the art you were doing, as you did it, rather than having it take over whole parts of the process as it is currently used.

I’m glad you’ve found value in it. I’ve played around with a similar workflow you describe in step 3 and 4, but I just find that it always produces the blandest version. Sometimes you get a surprising iteration, but I think being used to seeing visual patterns makes it much more obvious just how predictable the image details get. It just ends up looking like a compilation of techniques.

I’ve used it and continue to research it.

You’re wrong. You can’t have the same level of control as the person above describes. Even if you train the model on your own work, it will still be the one generating every “stroke” of the pencil. If trained for it, it will do it based on how you must often do so, but you can’t clearly control it. You can’t control granular details of the process of creating the image. It’s all broad strokes. I don’t know what your level of experience with art is, but so much of what makes art is tied up in the process of having to think through every little addition you make to an image. And by little addition I don’t mean “let’s add a person here” but “let’s do these 200 individual strokes that make up that person”. The involvement in the process is the point, and when an image is generated for you, you remove so much of the involvement and granular decision making, that the actual point is lost.

It’s like cooking with premade, pre-prepared ingredients. You can pick the dish and put it together with the stuff you buy, but you can’t control the whole process, because you’ve given up that for the sake of speed and convenience, and the dish will be different for it.

Yep. I see it a bit like asbestos. It’s being added to everything as the new cure-all to every problem imaginable. And similarly I think we’ll see some pretty rough repercussions down the line.

Ah yes, those sour grapes, indeed.

I disagree with your analogy, as I find it overstates the active involvement of the driver (prompter) during the drive (actual image generation).

Preparation is it’s own process, whether you’re curating art you made yourself/stole from non-consensual artists, or have been finding references as an artist. Different skillset. They help the process of making the final image, but they are not a direct part of that process.

And let’s not kid ourselves about theses datasets. There’s no accountability so there’s no way to ensure that any dataset you’re getting from other people aren’t comprised of, at least partially, stolen art.

ControlNet let’s you add visuals to your prompt for greater control, but you’re still generating the image externally, and leaving the vast majority of the decision making to the model you’re using. Even if someone is happy with the result they get from a generative model and find it visually pleasant, that doesn’t make it art. The model is doing the work and the model cannot have artistic intent, so it cannot make art. It can make images and people can enjoy those, but those images aren’t something new.

They are amalgamations of most basic common denominator of existing things. It is much more like a really advanced collage that is great at hiding the seams.

That is their perogative. It’s still antithetical to the whole concept of art, and if they sell AI generated images as art, then they are no longer artists, but just a middleman for generative images. Whether those images were trained on art or not.

AI art is also really prone to breaking when fed AI generated images, so it needs artists to work, but it’s use in the industry devalues the artists labour by being able to flood the market with low value replacements for art, thereby pushing actual artists out of the market and it’s own training pool. If the art industry cannot support professional artist because they are driven out of the industry by falling wages, then there will almost only be AI images left, accelerating the staleness of AI generated visuals.

Artists intent makes the artist, not the ability to make images. Otherwise art would have ceased to exist when cameras got sufficiently capable.

I understand the excitement, but it is very much a situation of a layman trying to describe to experts what the expert and all their peers need.

I think it is just because AI has been hyped so much, and has genuinely made such impressive progress that people get swept up by the excitement, and idea that they could make their ideas into something tangible. They just don’t know the amount of consideration that goes into translating that.

Right now AI art is like Google translate poems.

Right now it is not a tool. Right now it is an attempt at replacing artists.

It could be implemented in existing softwares in parts to make it a useful tool. Like a tool that could easily recolour parts of a fully rendered illustration, while still respecting the artistic intent with the form and lightning.

But right now it just spits out the blandest stuff, based on what it has identified as the most common denominators in art.

I strongly disagree. 99% of the work is being done by an algorithm. It’s like if we had autonomous driving and you said you were actively driving all day, because you told the car where to go, and then took a nap in the car until you had arrived.

AI art is antithetical to art. Art requires artistic intent.

It could have some limited application for very early exploration in commercial art, or perhaps as very limited tools used in existing art software, but generative art is inherently pointless and you need artists to be able to do incremental iterations properly, which is required for real work, which isn’t supported yet. I’ll sure it’ll get better and more convincing, but it’s still inherently pointless to use AI for art, since the is supposed to be human expression.