Spaces:

stabilityai

/

stable-diffusion

like 10.8k

Running on CPU Upgrade

Very interesting results. Would like to know the prompts that resulted in these.

For example, the third one's pormpt:
Fantastic sanctuary landscape of the elegant midnight, beautiful, sacred, cool blue water spring and Greek Temple in mysterious Cave, magnificent color matching, cinematic moody blue moon lighting, shinning effect, realistic painting, concept art, trending on art-station, Final Fantasy, Unreal Engine, Playstation5, long exposure, ultra-HD, 8K

Wow, that's a long string. I haven't gone that far yet.

This will be an unpopular opinion, but...

If your prompts are very general and don't point to a specific artist (or group of artists) and don't reference anyone else's IP, then I don't think you have anything to worry about.

However, if you specifically reference a single or a relatively small group of artists (for example, those whose work is trending on Artstation), or you reference IP like Final Fantasy, then, yeah, some people will call it theft. If you were to look closely through the works that were trending on Artstation when the bot crawled the net for image sources, you might find elements in your generated images that clearly originate from a particular piece of art. But I also think that SD does a pretty good job of combining multiple sources, so I doubt you'd see anything that looks like it was taken directly.

And, as for theft, well, who was it that said, "Good artists borrow; great artists steal"? Picasso. Same guy who popularized collage. And I think SD is ultimately just a very high tech form of collage.

If you're posting generated images online, I'd suggest that if you really want to have peace of mind, make sure people know they are AI generated. And if an artist claims your image looks too much like something they created the old fashioned way, be gracious.

Thank you Moriad, I almost agree with you.
Also, I've remembered an essay written about copying. A bit long, and I could not find the original source, but I'd like to share.

"In music or writing we don’t expect the child to burst naturally into creativity without first having gone through the laborious process of mastering the basic techniques, and this often involves copying what the teacher does, whether it be playing a particular chord or correctly writing the letter “b.” In fact, there is usually some element of copying involved in almost everything we have to learn. It is odd, then, that in drawing young children are expected to develop without help in this way. Apart from providing the materials and perhaps suggesting and discussing a stimulating topic, many parents or teachers rarely give further help. They see their role in terms of providing encouragement but not of giving direct assistance in the drawing process itself. The child is expected to find her own way and this is what creativity is supposed to be all about. Not surprisingly, then, adults often express unease if children copy the way other people draw things, especially if they copy the characters from books or comics.
In spite of these adult doubts, however, children often do copy drawings, greatest artists have routinely copied the work of the great masters who went before them. Understanding how others have achieved a certain effect hasn’t prevented them from moving on and trying something different. Nevertheless, you say, it may prevent ordinary people from doing so. But I suspect that without the experience of copying, they will not only fail to move on but they will not have got very far, artistically, in the first place. Although I would certainly not wish to reduce drawing to the level of copying alone, I do think that we can use children’s copying ability in positive and constructive ways. By disapproving of their copying activities, we may be closing the door on a very useful way of maintaining their interest in drawing and in widening their knowledge of the many ways in which things can be drawn. Far from having a dulling effect on creativity, copying can be used to open it up."

I don't think SD is an artist but a tool, very very strong tool. I believe creator or artist is a human who wrote the prompt, because you need knowledge and skills to write them, and it takes time. Certainly, SD is unprescedent high-tech tool to generate paintings or pictures. However, without users, SD wouln't generate any works.
I hope the works generated by AI would be a part of human arts. They are based on the data of works created by human artists, and AI works could also perhaps inspire people like real artists' works.

However, I have a second question.

Will Image-Generating AI threaten Human Artists?

BauBau, oh, I totally agree. Copying others' works is one of the best ways to master specific techniques, whether that's drawing, painting, digital art, or even writing. People who tell others not to copy are being silly. The only time it matters is when you're presenting the work as your own. If you paint a copy of one of Monet's waterlilies, it's your painting and you can sign your name to it. But the title should include something like "... after Monet." Most of his stuff isn't under copyright, but not owning up to your reference would get you laughed at, I'd think. Works that are under copyright are a different story, but there's nothing wrong with copying them for practice or to decorate your walls.

I mentioned collage because it's considered original work, but in its most recognizable form, it contains elements taken directly from others' work. Magazine clippings pasted together, for example.

Is AI art a threat to human artists?

Did cameras kill portrait painting? No, but they probably put some people out of work.

At the moment, the AI's are way too dumb to put anyone out of work, in my opinion. Even the best ones still don't always know how many fingers a human hand is supposed to have. Put a group of people close together, with their arms around each others' shoulders, say, and the AI doesn't know whose hand or arm or leg belongs to whom, which can cause all sorts of gruesome problems.

Now imagine you're doing a commission, and the client says, "I love it! I want it EXACTLY like that! But can you move the car about 2 inches down?" Hmmm. What's the prompt for that?

Sure, it will get better. But have you watched a pro illustrator at work? They are fast. No generating a thousand images to get kinda what you want, except the wheels aren't actually round, and the driver of the car has three arms.

Unless you hire Boris Vallejo, that is. 😄 There's a picture somewhere that he and Julie Bell did together, and one of the women in it has either three arms or three legs, can't remember which. So if humans can mess up when they decide to entangle several people together, it's not going to be easy to get an AI to get it right reliably.

I guess we'll find out.

You can spend a full weekend on creating a drawing using an AI. Tweaking prompts, selecting seeds you like, copy & pasting and letting img2img blend stuff.
In my opinion a big part of creativity is selecting stuff you like.
If I code up some generative art algorithm (like I sometimes do), a big part of the process of releasing a finished image is the selection process and tweaking values of that algorithm. Same if you produce music using a hardware modular synthesizer. You can plug in complex signal modulations and create a beautiful ambient piece with relative low effort if you know your tool a bit. Record it, push it to bandcamp and become an ambient musician.
Or put it this way: Is a musician only someone who plays an instrument? Is someone who clicks in a few notes in a DAW, loads plugin presets a music producer? Are they a music producer only after they tweaked their musicial piece for an hour? a weekend? a week? a month?
Taste is the key, what drives your selection process. The rest is learning and knowing your tools. The hard part with AI is going to be producing prompts that make the AI do what you like, and partially waiting for the right seed, and maybe make a collage of AI generated art pieces.

Someone pointed out this article to me recently: https://andys.page/posts/how-to-draw/#
The result is awesome, and required someone sitting there for a weekend and investing their time for a weekend.

The tools change what is possible, like someone pointed out cameras already. You can use a camera creatively, or you can choose to just record what you are seeing. Like scanning a document. Or you can get up at 4am, drive 2 hours to that special location on a certain day and catch just the right light to shoot a single perfect picture.

I'm a software developer, just because the Github Copilot AI makes parts of the programming process easier will not put a software developer out of business (I'm not going into the big licensing debate here about the code base that was used to train the AI). Software development is way more than just writing down code. It's also knowing what kind of code you need in the first place, what system architecture will suit the requirements, and how to design your codebase to be maintainable.

If a young children put in some prompt, and took 5 seconds to choose a result and has the fun of their life. And if they publish it and it becomes the next big art thing, that is fine in my opinion. There is an artist who just put a piece of butter somewhere and that is called art... https://en.wikipedia.org/wiki/Fettecke

If putting a piece of fat somewhere is not a low bar for art (and creativity), I don't know what is.

Thank you so much WeiCon, your opinion is very persuasive, and the article you suggested is interesting too.
I'm going to sophisticate my prompts and create enjoyable works!
I think there is a large space to modify and improve the accuracy of prompt to make the AI what I imagine.

I appreciate what you told!

Recently an AI won a prize at an Art contest, i think it's just like chess.
Do you want perfect play, play a chess bot, and next year there will be a better bot
Do you want to see the best human play follow Carlson
Or do you want to enjoy chess, join a chess club, play with friends etc?

What do you give value, ea lot svg graphics cost money, well they have been made...but the artistic quality is incredibly low (for a real artist); though they sell.
Do you need some fast graphics low budget for your logo...you're not going to ask Davinci (if he was still alive).
At some point, one realizes the true skill, and handcraft, of the real human artist.
I paint art myself, I think it's funny what these image generators do, but receiving or giving a real hand-painted means a bit more to me (and if a robot did it?..in some sense less )

But what if it gets so good that I can fool people to be a lost Rembrandt?

There have been humans that forged artworks of famous artists, see Beltracchi: https://en.wikipedia.org/wiki/Wolfgang_Beltracchi
AIs can't yet hold a brush or mix colors, so there is still a way to go to make a painting forgery AI.
AIs don't defraud people, people defraud people.

And I believe, once AI artworks are ubiquitous, presenting great paintings that were made by a human will still be amazing.
Heck, maybe AI inspires humans to try new art styles or mix art styles. There is a whole lot of inspiration to gain from AI.
Just like Chess and Go playing AIs have become the teachers and inspiration for professional players.
Have Chess and Go engines used as cheating device? Heck yes, in online and offline competitions. One instance of (trying to) pulling that off in off line I read about just the other day: https://incoherency.co.uk/blog/stories/sockfish.html
There has also been opened a whole positive world for humans though: Go students now have a professional player at home that analyzes their games, this is amazing, and it will and did already improve the world of human Go playing by a lot. With SD, artists now have an amazing (sometimes a bit iffy though) artist at home, as source of inspiration and as it's own tool for new art.

In my opinion, Ai isn't really a threat to actual artists because generations are often unpolished and things are usually a little off.
Perhaps a concept artist might need to find a new job though! 🤣

PS Its not only about painting artists

I was just thinking, how about fashion... that is a multi-million industry.
What if it becomes so cheap to do, that Amazon 'dreams' your clothing based upon a photo of you?
Next, a robotic industry creates your clothing that way...
This becomes possible pretty soon (a few months at most, at max before 2024).

What kind of effect will that have on people? society?
Already the TikTok generation becomes addicted to an algorithm that picks the next videos you most likely like too.
Could such tools in the hand of industry become addictively dangerous? ea; get rich as the next Elon Musk in fashion?
Are we maybe wasting time and effort, as we can already make Art and make fashion the old way?
Could such processor power and human labor be better spent?

Economic fashion singularity event ahead...

I am starting to feel that in a practical manner when assessing SD that it indeed is a 'creative' generator. Let's assume that creativity boils down to coming up with new / original ideas. These ideas can be visual (like a painting), or complex abstract ones (like Einstein's theorem),
they relate to each other in that ideas require a (high) level of creativity. To think outside of the box, as the cliche dictates.
However, there's this quote from Mark Twain about ideas that really captures the essence (of the creative process).

“There is no such thing as a new idea. It is impossible. We simply take a lot of old ideas and put them into a sort of mental kaleidoscope. We give them a turn and they make new and curious combinations. We keep on turning and making new combinations indefinitely; but they are the same old pieces of colored glass that have been in use through all the ages.”

Stable Diffusion's dataset is the 'collection of old ideas' while itself being an automated / programmed 'mental kaleidoscope'. Just like us humans, we can only create by combining the concepts we have learned, the same is true for SD's output.

Art history shows us that styles, their form and shape merge with other styles from around the same time. Taking old ideas and finding an original configuration that feels new is beautifully analogue to finding values in the 'latent space' like Stable Diffusion does.

The picture I'm (pardon my phrasing) painting here kinda suggests why some people are more creative than others. By being open to new experiences, an open mind takes in more "foreign" ways of seeing the world, expanding their dataset and latent space where new ideas can be sampled from.

Long story short: Yes, SD is a model that has creativity.

Yes, you are expressing creativity by writing a prompt and making a selection from the outputs. It's not the same as painting it, or drawing it, of course.

No, SD AI is not theft of creativity. Creativity is the child of old ideas. Artists who display their work in public can't expect to not be mixed into future ideas of others, it's just silly.

Oh man, what a world.

Is AI art theft of creativity?

i've created art for YEARS - pen, pencil, paint on canvass, creating 3d models, rendering in blender or daz studio - i prefer the computer because it's not messy like paint is and i can create what I'm actually visualzing.
I have to be far more creative to come up with a prompt that will get the AI to make what I want, than I have to with any other medium. AI is just a tool - if you choose to be lazy about it, it'll give you something, sure. But if you have a vision, and you work with the tool till that vision comes to life, it's not the AI doing that, it's you. Just like it's you using that pencil to put lines on paper and erasing holes in said paper as you make changes.

But I can make many kinds of paintings by AI.
Is this creativity?

sure - becaue you are thinking about what to tell the AI to do and using it to do that, and exploring inside you at the same time. Ever tried generating an image without giving it a prompt at all? You get an error back. YOU have to input something or tool just sits there and does nothing.

It's fun to read most of us in this thread seam to be artist as well we paint sculpt or blender, and found this software.

It would be quite easy to create some random text (for example use bloom or gpt2 or gpt3) then create some art.

As for now though it's 512 or 1024 pixels. I can use it for inspiration my paintings are a lot larger let's keep it small like this.

Artists who display their work in public can't expect to not be mixed into future ideas of others, it's just silly.

1. greg's missing a golden marketing opportunity - people know him now who would never have heard of him and a lot of those people would like an authentic piece of art created by the man himself. Instead, he's griping to the media and alienating a lot of potential customers

2. artists have always used other artists for inspiration - and copied their styles - it's how the human mind works. This isn't new, and it won't ever stop happening.

@PeterB123A

Upscaling is a thing =P

Yes, upscaling is a thing. And if you're not using this upscaler https://www.imgupscaler.com/ check it out. up to 400% and works extremely well. then run the upscaled image if it has a face through GFPGan, and possibly also photoshop if needed.