Google takes down Gemini AI image generator. Here’s what you need to know.

[ralfy]

That "Founding Fathers" image is pretty funny considering the ideas of being an American, Freedom, Equality and all the other hypocritical stuff was never intended to be extended to the slaves working in American fields.

It gets better:

Google takes down Gemini AI image generator. Here’s what you need to know.

The human race, or for that matter the Earth, one day, of nature's own astrological due course, cease to exist (and they will), the universe will simply shrug. That said, there's no reason in the heavens or Earth, to hurry that along any more than it needs to be. True, which is why...

debatepolitics.com

 

[Infinite Chaos]

Of course, i don't understand who is hurt by ai generated images of a black founding father or a female pope... that will take a bit of explaining.

Firstly, the illustrator who would have been paid to produce the image and secondly, all the photographers and artists whose work has been used without their permission to "train*" the AI.

Personally I see three trends for artists and illustrators

• Give up learning the skills of being an artist or illustrator and use the tool instead to produce the work. Most artists / illustrators would take at least a week to produce the original work (from research / testing their materials and media to find what works) but an AI can produce a high quality blend image in 60 seconds for a cheap monthly subscription price.
• Stop posting their work online - if it's not online for an AI to steal from, your work is safe and AI image generators will continue to produce the same type of images / the audience gets bored and eventually wants something new. You can already see that now in many of the images produced by different AI. The colours / shading / poses and types of character are all actually quite similar and Google's generator now is producing very similar images to what I saw when MidJourney was first making waves in the industry. MidJourney hasn't moved on either - maybe faster, maybe the language of prompts has evolved to the maximum and stylistic "originality" is universal.
• Somehow learn to incorporate the technology into their work - except the AI will always produce work faster than any illustrator or artist could ever do.

* The word "train" doesn't really fool artists or illustrators, the original work is stolen and blended together to create a soup approximating what the user requests.

 

[OlNate]

Firstly, the illustrator who would have been paid to produce the image

Probably not a lot of pay for images of female popes and black founding fathers, but i could be wrong.

and secondly, all the photographers and artists whose work has been used without their permission to "train*" the AI.

That's an interesting one... but in the current state, if I'm an artist, i can easily look up the same content to inspire my own work, free of charge.

Personally I see three trends for artists and illustrators

• Give up learning the skills of being an artist or illustrator and use the tool instead to produce the work. Most artists / illustrators would take at least a week to produce the original work (from research / testing their materials and media to find what works) but an AI can produce a high quality blend image in 60 seconds for a cheap monthly subscription price.

For commercial art, perhaps, but even there, you still need to understand art to be able to generate it well using AI prompts, so they'll need to know art theory at a minimum to produce decent work.

• Stop posting their work online - if it's not online for an AI to steal from, your work is safe and AI image generators will continue to produce the same type of images / the audience gets bored and eventually wants something new. You can already see that now in many of the images produced by different AI. The colours / shading / poses and types of character are all actually quite similar and Google's generator now is producing very similar images to what I saw when MidJourney was first making waves in the industry. MidJourney hasn't moved on either - maybe faster, maybe the language of prompts has evolved to the maximum and stylistic "originality" is universal.

Again, most art being affected by AI would be for commercial use, and therefore would most likely need to be posted online at some point. Also, since the art belongs to the customer, the artist wouldn't have any say.

in terms of styles, there is a ton of content creators that make plug ins for AI image generators that address this problem. You can browse massive collections of these plug ins, many of them free, customizing everything from the model you want to incorporate into the image to the entire world your image takes place in.

And, of course, prompt engineering is a big element as well - how you're able to communicate your ideas to the machine. If you're not good at this, it is very difficult to get what you want from AI, and the results can be really weird and unusable.

• Somehow learn to incorporate the technology into their work - except the AI will always produce work faster than any illustrator or artist could ever do.

This is the big one, yes. But, again, the product is heavily dependant on the prompts the user gives the AI, so it's an easier said than done kind of thing.

This isn't a problem, though. The need for digital content continues to rapidly climb, told like this are required to keep up. Every graphic designer i know, and i know a few, is using some element of AI in their work because they are too busy to do it all manually. And, once they get over the kind of concerns you're bringing up here, they love it - its an instant raise, in terms of what they can make in an hour. Same for copy writers, web designers, SEO specialists, video editors and video content creators.. and that's just within the digital marketing industry.

* The word "train" doesn't really fool artists or illustrators, the original work is stolen and blended together to create a soup approximating what the user requests.

Which is exactly how a lot of human generated commercial art gets done by humans. I have never really understood this concern. I guess the only difference is that with AI is that you can theoretically track where it got its inspiration from, and the company creating is a better target for lawsuits given their deeper pockets.

It's true, the way this work gets done is definitely changing, and i think it's important to have these discussions to ensure that we bake some ethics into it. But i also think there is a tendency to assume the worst case scenario is the only scenario when faced with change, and that's not always accurate. It's important to partition the "Chicken Little" stuff from the real concerns, in order to make that process as efficient and effective as possible.

 

[Infinite Chaos]

Probably not a lot of pay for images of female popes and black founding fathers, but i could be wrong.

An illustrator charges an hourly or per peice charge whether it's a female pope, a family eating at dinner, a cycling chimpanzee or a black founding father.

but in the current state, if I'm an artist, i can easily look up the same content to inspire my own work, free of charge.

True, but what makes a human piece of interpreted work is that the artist has not simply copied works and made a "soup" of copied works. There is always (in better works) be effort that uses inspiration and not duplication.
It's one of the oldest debates any art teacher has with their new students and one I had for 25 years with new artists / illustrators and designers - that they wanted to produce original work but the tradition in art school is to be inspired, "take inspiration" but not simply duplicate the works they are seeing.

How to develop your ideas in an Art project

A level art students must present a Coursework portfolio that shows development. What does development mean? This article explains.

www.studentartguide.com

Scroll down to number 4. That's the human interpretation part. An AI doesn't "interpret" - it takes cue words, finds art or images that link to that and then create the nearest approximation of the original work.

For commercial art, perhaps, but even there, you still need to understand art to be able to generate it well using AI prompts, so they'll need to know art theory at a minimum to produce decent work.

I taught a photoshop night class to business owners for 3 years in my last few years as an art teacher. Those business people were all users of a website called "Fiverr" where you could take your business and get some Filipino artist to produce your work for $15 as opposed to paying $500-$600 to a trained Western Artist. They were all looking to see how they could remove that $15 payment from their costs.

I guarantee you, those same people and their type will be paying the small subscription fee to produce work and they are not bothered with the training of art theory or anything I used to teach to produce that work.

Again, most art being affected by AI would be for commercial use, and therefore would most likely need to be posted online at some point. Also, since the art belongs to the customer, the artist wouldn't have any say.

You're not following me. Even with your graphic design friends who are using AI art at the moment, they are simply using a clever tool that regurgitates what is already online. Nothing of what is being produced is "new."
For the moment, it's shiny, quick and all the buzz but as I said in my previous - I certainly have spotted similarities in the output of those works. For the first period of time, artists like Greg Rutkowski were seeing images heavily based on his style flood sites like MidJourney.

AI: Digital artist's work copied more times than Picasso

Greg Rutkowski is among the artists calling for more protection from artificial intelligence tools.

www.bbc.co.uk

If I were a budding digital Greg Rutkowski right now, I would be very reticent of posting my best portfolio online

The need for digital content continues to rapidly climb, told like this are required to keep up. Every graphic designer i know, and i know a few, is using some element of AI in their work because they are too busy to do it all manually.

See above. The law of diminishing returns will eventually come through - unless artists and illustrators start to use tools like Nightshade which prevent AI stealing and copying their work.

I have been talking to many of my ex students about this and promoting Nightshade or similar.

Which is exactly how a lot of human generated commercial art gets done by humans. I have never really understood this concern.

No it's not.

It's true, the way this work gets done is definitely changing, and i think it's important to have these discussions to ensure that we bake some ethics into it. But i also think there is a tendency to assume the worst case scenario is the only scenario when faced with change, and that's not always accurate. It's important to partition the "Chicken Little" stuff from the real concerns, in order to make that process as efficient and effective as possible.

Without image scramblers designed to destroy what AI does, there is no future. Ethics went out the window last January when the technology exploded onto the world.

 

[OlNate]

An illustrator charges an hourly or per peice charge whether it's a female pope, a family eating at dinner, a cycling chimpanzee or a black founding father.



True, but what makes a human piece of interpreted work is that the artist has not simply copied works and made a "soup" of copied works. There is always (in better works) be effort that uses inspiration and not duplication.
It's one of the oldest debates any art teacher has with their new students and one I had for 25 years with new artists / illustrators and designers - that they wanted to produce original work but the tradition in art school is to be inspired, "take inspiration" but not simply duplicate the works they are seeing.

How to develop your ideas in an Art project

A level art students must present a Coursework portfolio that shows development. What does development mean? This article explains.

www.studentartguide.com

Scroll down to number 4. That's the human interpretation part. An AI doesn't "interpret" - it takes cue words, finds art or images that link to that and then create the nearest approximation of the original work.



I taught a photoshop night class to business owners for 3 years in my last few years as an art teacher. Those business people were all users of a website called "Fiverr" where you could take your business and get some Filipino artist to produce your work for $15 as opposed to paying $500-$600 to a trained Western Artist. They were all looking to see how they could remove that $15 payment from their costs.

I guarantee you, those same people and their type will be paying the small subscription fee to produce work and they are not bothered with the training of art theory or anything I used to teach to produce that work.



You're not following me. Even with your graphic design friends who are using AI art at the moment, they are simply using a clever tool that regurgitates what is already online. Nothing of what is being produced is "new."
For the moment, it's shiny, quick and all the buzz but as I said in my previous - I certainly have spotted similarities in the output of those works. For the first period of time, artists like Greg Rutkowski were seeing images heavily based on his style flood sites like MidJourney.

AI: Digital artist's work copied more times than Picasso

Greg Rutkowski is among the artists calling for more protection from artificial intelligence tools.

www.bbc.co.uk

If I were a budding digital Greg Rutkowski right now, I would be very reticent of posting my best portfolio online



See above. The law of diminishing returns will eventually come through - unless artists and illustrators start to use tools like Nightshade which prevent AI stealing and copying their work.

I have been talking to many of my ex students about this and promoting Nightshade or similar.



No it's not.



Without image scramblers designed to destroy what AI does, there is no future. Ethics went out the window last January when the technology exploded onto the world.

Just a small question before i dig into this. Did copyright laws get repealed without me knowing about it? It feels like at least some of your concerns around they would be addressed by existing law, no?

 

[Infinite Chaos]

Just a small question before i dig into this. Did copyright laws get repealed without me knowing about it? It feels like at least some of your concerns around they would be addressed by existing law, no?

Haha, one of the first things I learned when I was doing my Design degree (we did a whole module on laws, copyright and how to protect yourself) was that copyright law is only as good as the amount of money you have to enforce your rights.

I can't remember any of the examples demonstrated to me in 1984 but the message was rammed home that copyright law only goes so far.

 

[OlNate]

Haha, one of the first things I learned when I was doing my Design degree (we did a whole module on laws, copyright and how to protect yourself) was that copyright law is only as good as the amount of money you have to enforce your rights.

I can't remember any of the examples demonstrated to me in 1984 but the message was rammed home that copyright law only goes so far.

I see. So people could steal before AI.

I dunno, bud, i guess we see this differently, and that's OK, not every chat is going to end in consensus. Having literally just finished a digital marketing program (my mid life crisis involved a new career rather than a new car or woman lol), i see the opportunity rather than the threat... but i haven't been doing things another way since 84.

I guess we'll just have to see. In the meantime, if nothing else, I'm trying to get my son interested in it, hoping he'll want to have a career in building it. It'll be the last job done by humans, so at least the job security factor is there.

 

[Infinite Chaos]

I see. So people could steal before AI.

Yep, however as I said before, your pocket/wallet size dictated success rates. "Borrowing from" or "stealing from" is a standard part of Art & Design once you get students to realise nothing is truly original but what we focus on is that they make the effort to personalise what they are taking from.
If you looked at the previous link to an exam paper, you'll see that students are encouraged to do several copies and variants. It's not to get better at making duplicates but it's to start to personalise the new work so it isn't an obvious copy.

It's called "using influence." Picasso's cubism stage wouldn't have happened if he hadn't been invited to a private viewing of African tribal masks but he didn't simply copy or duplicate the masks - he made something of his own from it.
That is what AI cannot do - it simply makes blends in the hope that the viewer is easily satisfied. That's why Getty Images have such a strong case against Stable Diffusion.

I take it you saw some of the early scraped images?

i see the opportunity rather than the threat

True, there is opportunity. Number one - why bother putting yourself through 4-6 years of art college and developing skills when all you need is to write a few prompts, reference an artist whose style you are after and can wait 60 seconds?

I guess we'll just have to see. In the meantime, if nothing else, I'm trying to get my son interested in it, hoping he'll want to have a career in building it. It'll be the last job done by humans, so at least the job security factor is there.

In AI?

 

[OlNate]

Yep, however as I said before, your pocket/wallet size dictated success rates. "Borrowing from" or "stealing from" is a standard part of Art & Design once you get students to realise nothing is truly original but what we focus on is that they make the effort to personalise what they are taking from.

I think the same can be said about AI, though.

And from a practical point of view, that's kind of how you want it. Imagine if Pepsi came out with a polar bear ad the year after Coke did it? Anything worth stealing would not be worth stealing.

If you looked at the previous link to an exam paper, you'll see that students are encouraged to do several copies and variants. It's not to get better at making duplicates but it's to start to personalise the new work so it isn't an obvious copy.

It's called "using influence." Picasso's cubism stage wouldn't have happened if he hadn't been invited to a private viewing of African tribal masks but he didn't simply copy or duplicate the masks - he made something of his own from it.
That is what AI cannot do - it simply makes blends in the hope that the viewer is easily satisfied. That's why Getty Images have such a strong case against Stable Diffusion.

Again, sorry, i gotta disagree. The prompts are the personal spin. The AI needs input, and that input still requires creativity.

I take it you saw some of the early scraped images?

lol... yeah, you can get some crazy outputs if you don't know how to craft good prompts..

True, there is opportunity. Number one - why bother putting yourself through 4-6 years of art college and developing skills when all you need is to write a few prompts, reference an artist whose style you are after and can wait 60 seconds?

Why indeed! Imagine what could arise if creativity was not limited by aptitude. I can't draw beyond doodles, but i can picture things I'd love to if a had the talent. What is the downside of giving me the ability to share that creativity without taking 4-6 years worth of schooling? That seems like gate keeping to me, i can't get behind that.

In AI?

Yep. It's the last job to get replaced.

 

[Infinite Chaos]

I can't draw beyond doodles, but i can picture things I'd love to if a had the talent. What is the downside of giving me the ability to share that creativity without taking 4-6 years worth of schooling?

That's the core of why we'll never agree on this.

Thanks however for the discussion.

 

[OlNate]

That's the core of why we'll never agree on this.

Thanks however for the discussion.

OK, no worries... i think i understand the issue at this point anyway.

And thank you for the discussion as well!

 

[Infinite Chaos]

OK, no worries... i think i understand the issue at this point anyway.

And thank you for the discussion as well!

A friend posted this to me today. I had pointed out a few months ago that Ai images had very quickly started to have a strong "look alike" feeling to them. AI experts have been calling this "convergence" apparently; AI takes images from the web and uses them to "train" and create more images.

All the time, the pool of images will become less as fewer people will either make original work or stop posting online and you are left with AI using AI references for new work.



Obviously, that makes me happy but I am aware it is a selfish view as an artist.

 

[OlNate]

A friend posted this to me today. I had pointed out a few months ago that Ai images had very quickly started to have a strong "look alike" feeling to them. AI experts have been calling this "convergence" apparently; AI takes images from the web and uses them to "train" and create more images.

All the time, the pool of images will become less as fewer people will either make original work or stop posting online and you are left with AI using AI references for new work.



Obviously, that makes me happy but I am aware it is a selfish view as an artist.

Not to wreck your day, but check this out: https://civitai.com/

 

[Infinite Chaos]

Not to wreck your day, but check this out: https://civitai.com/

Don't worry, doesn't wreck my day at all. I googled the ai and this article on Reddit popped up:



One comment out of many that reinforced what I previously wrote says

"It turns out that even with the most advanced art tool ever created most people who are not artists aren't actually very creative.

All we get is an endless supply of mainly female characters in various themes and a massive amount of anime girls.

Most of the checkpoint models are merges because training models from scratch is time consuming and costs money. That means there's going to be very little difference between a lot of models.

Smaller models like Loras are driven by what the bulk of the Civitai users want and that's anime girls, female celebs and porn. If you look at something like Loras with all the filters off you will see it's probably about 98% anime girls and porn related content uploaded everyday. Embeddings aren't much different, they are dominated by female celebs and Instagram girls."

That is a depressing future if human artists give up making work. I have no clue what checkpoints, embeddings and Loras are or mean.

 

[OlNate]

Don't worry, doesn't wreck my day at all. I googled the ai and this article on Reddit popped up:



One comment out of many that reinforced what I previously wrote says

"It turns out that even with the most advanced art tool ever created most people who are not artists aren't actually very creative.

All we get is an endless supply of mainly female characters in various themes and a massive amount of anime girls.

Most of the checkpoint models are merges because training models from scratch is time consuming and costs money. That means there's going to be very little difference between a lot of models.

Smaller models like Loras are driven by what the bulk of the Civitai users want and that's anime girls, female celebs and porn. If you look at something like Loras with all the filters off you will see it's probably about 98% anime girls and porn related content uploaded everyday. Embeddings aren't much different, they are dominated by female celebs and Instagram girls."

That is a depressing future if human artists give up making work. I have no clue what checkpoints, embeddings and Loras are or mean.

Hehe... I'd argue, but i won't... I like this conclusion: nothing to worry about.

 

[ralfy]

 

[joluoto]

The funniest AI thing is still Will Smith eating spaghetti:

 

[ralfy]