Protecting Your Creative Works From Data Scraping [Mod+]

[Art Waring]

This is a [+] thread, specifically addressing ways for creatives to protect themselves from harmful data scraping and the theft of their intellectual property. If you are pro-ai, there are plenty of other threads for you to talk about that, while this thread is specifically for creatives and publishers looking to protect themselves from the inherent risks of gen-ai. This is not another place on the internet for techbros' to bash on artists that are simply trying to earn a living.

Anyway,

Here is a link to a short article about how creatives can protect their work from the invasive technology that is generative-ai.

Creatives are voicing concerns over protecting the rights to their work in light of the looming capabilities of [Generative] artificial intelligence (AI). These anxieties aren’t unfounded, as artists continue to fight copyright lawsuits against Midjourney and StabilityAI for training their AI tech on creative content without the consent of the artists.

“AI steals art and generates off of it,” Pittsburgh-based animator and studio artist Alexis Raine said. “AI is already impacting the art industry.”

Whether you’re a writer, a photographer, or a visual artist, you deserve attribution and compensation for your original works and efforts. Here are a few ways to protect your work from dreaded AI scraping.

TLR: Glaze & Nightshade are a thing. As well as a new notice you can put into your publications to prevent data scraping.

I know that the techbro's on the forums like to bash on Glaze/ Nightshade saying it isn't effective (well good for you techbros, bragging about how easy it is to steal work from people who are struggling to pay the rent), but it is still in development and will eventually become more effective (we hope so, anyway).

Here is the new written notice you might want to include in your publications:

“NO AI TRAINING: Without in any way limiting the author’s [and publisher’s] exclusive rights under copyright, any use of this publication to “train” generative artificial intelligence (AI) technologies to generate text is expressly prohibited. The author reserves all rights to license uses of this work for generative AI training and development of machine learning language models.”

I updated the notice myself to include images, as they were absent from the original statement, as it was created to protect writers, while visual artists might want to include this notice instead:

NO AI TRAINING: Without in any way limiting the author’s [and publisher’s] exclusive rights under copyright, any use of this publication to “train” generative artificial intelligence (AI) technologies to generate text or images is expressly prohibited. The author reserves all rights to license uses of this work for generative AI training and development of machine learning language models.

I am not a lawyer, I am not providing any legal advice, I am simply trying to help inform creatives about the tools currently in development aimed at preventing the theft of their creative works.

Please feel free to discuss anything like Glaze, Nightshade, or the crucial need for transparency in the face of generative ai in this thread (or any other issues affecting artists like the loss of work or the difficulties being faced by artists due to gen-ai).

This thread is not for the discussion of how neat gen-ai images generators are, or your personal hot take on the legality of gen-ai, as that is already being discussed elsewhere in great numbers. There is currently very little regarding information for creatives and publishers with concerns on the subject, and I would hope that regardless of your personal opinion, that you can respect that those of us who rely on our creative work to earn a living need some kind of place to discuss our concerns without being harassed or cyberstalked like I have seen from techbros elsewhere on the forums.

Respectfully,
Art

 

[Epic Meepo]

Thanks for posting this. I'd been idly contemplating the idea of something like Nightshade recently, but I hadn't gotten around to doing any research. I didn't know until I read the OP that Nightshade is already a thing. It's interesting to see someone's already exploring this idea.

To be honest, though, the team working on Nightshade seems to be underselling the real potential of this software. The stated goal of allowing human artists to data-poison their personal work is intriguing, but this technology would probably be more effective if it was used on a larger scale:

Instead of data-poisoning individual human artists' works as they get uploaded to the internet, Nightshade would be more effective if it was used by a bot (a counter-insurgent AI?) that generates and posts data-poisoned images to the internet a million times faster than any human artist could.

To put it another way, data-poisoning individual artists' work seems to me like missing the forest for the trees. To make a real impact on the development of AI technology, someone would need to use Nightshade technology to data-poison the internet itself, to the fullest extent possible.

 

[Art Waring]

Thanks for posting this. I'd been idly contemplating the idea of something like Nightshade recently, but I hadn't gotten around to doing any research. I didn't know until I read the OP that Nightshade is already a thing. It's interesting to see someone's already exploring this idea.

To be honest, though, the team working on Nightshade seems to be underselling the real potential of this software. The stated goal of allowing human artists to data-poison their personal work is intriguing, but this technology would probably be more effective if it was used on a larger scale:

Instead of data-poisoning individual human artists' works as they get uploaded to the internet, Nightshade would be more effective if it was used by a bot (a counter-insurgent AI?) that generates and posts data-poisoned images to the internet a million times faster than any human artist could.

To put it another way, data-poisoning individual artists' work seems to me like missing the forest for the trees. To make a real impact on the development of AI technology, someone would need to use Nightshade technology to data-poison the internet itself, to the fullest extent possible.

Those are some good thoughts on the subject, thank you for your contribution.

I think they are underselling Nightshade because it is still brand new, and will need further development in order to become even more effective.

I agree that more widespread measures will be needed, but in the meantime there are zero protections for artists, and this is the best we have ATM. For the moment, Nightshade (NS) is not "enough" in terms of achieving some kind of widespread poisoning of datasets, but what it will achieve is a gradual decrease in available artwork for data scrapers to vacuum up in the long term.

Currently scientists working on ai are predicting that gen-ai will run out of data to train on by as early as 2026. The constant need for new data means that eventually there won't be enough to train on (assuming artists adopt NS en masse). The more artists that use NS, the more that opt-out and protect their work, the less likely that they will be able to keep training models on stolen data.

Without artists to steal from, there will be no new models, and the current state of affairs will stagnate. Eventually gen-ai tools will become fodder for fiverr "artists," while real artists will continue to work once gen-ai can no longer steal their works wholesale.

The most ironic part of all of this, is OpenAI is worth 100 Billion dollars (yup, billions), with that worth propped up by gen-ai. They can afford to pay artists the world over for their work, but they are literally pleading poverty ("we can't afford to pay for training data, but we are worth billions").

 

[Scribe]

The most ironic part of all of this, is OpenAI is worth 100 Billion dollars (yup, billions), with that worth propped up by gen-ai. They can afford to pay artists the world over for their work, but they are literally pleading poverty ("we can't afford to pay for training data, but we are worth billions").

Nvidia, on the back of this "AI" boom, are now valued at over 3 Trillion.

I'll pour one out for the 'elites' who dont want AI in peoples hands...or something.

 

[Art Waring]

Nvidia, on the back of this "AI" boom, are now valued at over 3 Trillion.

I'll pour one out for the 'elites' who dont want AI in peoples hands...or something.

Yep, its pretty ridiculous.

 

[Art Waring]

Just an update for folks who don't yet have the hardware to use Glaze/ Nightshade, they have updated the tools to work with intel chips (they previously only worked on M1 or higher silicon chips), but I think that it takes more time to glaze your artwork.

Also, they have made "Webglaze" available to the public, which is a kind of public-service tool that allows you to use glaze if you can't run it on your own hardware. So they can glaze your work for you, but you do have to show your work (to prove you are not acting in bad faith) and get it approved. All in all, the tools are getting better, and in time I think things will be slowly improving for traditional artists.

Hope you have a good day, Cheers.
Art