Block robot scraping images

Forum|Forum|2 years ago
April 26, 2023
3 replies
1397 views

How does one block generative AI tools from using robot to scrape images from an Abobe Portfolio

Cloud storage web assets

This topic has been closed for replies.

Nancy OShea

Community Expert

If it's on the web, there is no technology on the planet that can stop people from taking your images if they have a mind to. Owing to how the web works, images are in the browser's memory and easily accessed. Or people can grab screenshots with the Print Screen key. You can't stop it.

The best you can do is heavily watermark low-resolution images to make them less desirable. Or use digital barcoding services to track how your images are being used online. Digital barcoding costs extra and embeds a barcode into the image file which is invisible to humans. There is a Photoshop Extension for it below. Barcoding won't prevent theft, but it will make it harder to get away with it. 🙂

https://helpx.adobe.com/photoshop/using/digimarc-copyright-protection.html

Regarding so called AI scraping software. That's a myth. Unless you're a very famous artist like Picasso or Monet, it's unlikely that anybody knows your work or cares about it. The only way AI could use your work is if a human manually trained AI to use it. In the absence of fame or notoriety, the chance of that happening is slim to none. Don't sweat the small stuff.

Nancy O'Shea— Product User & Community Expert

I

ianm48171564Author

Participating Frequently

Hi Nancy,

You are spot on about image theft on the web, and the near impossibility of preventing it.

You are correct when you discuss the use of an image as an inspiration (prompt) for subsequent AI generation of an image in the "style of a famous artist/photog" etc. My beef is with the image sets used to train the generative AI models using everyone's images as though they were in the public domain.

That is not my issue, but rather wishing that Adobe pursues an opt-out platform approach to making it possible to dis-incent the scrapers like those used to populate the LAION-5B dataset used to train generative AI tools like Midjourney and Stability.
Diffusion.

DeviantArt has taken this approach by including NOAI as a tag option, which, while not legally enforceable, is a good start, as it would violate its Terms of Service.
The watermarking option is useful for after-the-fact tracing, with Getty images, as you probably know, having initiated proceedings against Stability AI for scraping its images without a licence, based on watermark discovery.

I hope this clarifies my concerns.

Nancy OShea

Community Expert

Adobe has nothing to do with what generative AI services do. This goes way beyond their reach.

At the moment, Midjourney, Stable Diffusion and others are self-regulated which means they answer to no one. If so-called image scraping is to stop, the matter must be proved & hammered out in the courts. AFAIK, nobody has launched a civil or class action suit against any Generative AI image service and won. Until that happens, there is no conceivable way to regulate what diffusion serrvices do. But how do you put this genie back in the bottle? The body of artwork that's already in the ether can be reused ad infinitum.

There's no doubt that AI is operating in gray areas and using double-standards. On one hand, AI images are trained on artworks (public domain, original, copyrighted and non-copyrighted). A user can bring whatever they want into the mix. On the other hand, AI music services are afraid of legal entanglements with ASCAP and BMI. So they train their AI on music tracks that are in the public domain. It's cleaner that way. 🙂

Unfortunately, visual artists do not have the same protections that song writers have. And I think that's the first thing that needs to change.

Just my 2 cents.

Nancy O'Shea— Product User & Community Expert

kglad

Community Expert

@ianm48171564

what exactly are you trying to block?

I

ianm48171564Author

Participating Frequently

I'm not happy that any generative AI model scrapes all my images to add to their learning database at no cost to themselves .... specifically img2dataset robot ... furthermore I suspect Adobe would not be too keen on its portfolio servers getting swamped by the load.