Adobe Generative AI User Guidelines rule

[Encartauk]

So, I've just read through the terms using the link in the updated PS 2024:

"2. No AI/ML Training

When using our generative AI features, you agree you will use them only for your creative work product and not to train AI/ML models.

This means you must not and must not allow third parties to, use any content, data, output or other information received or derived from any generative AI features, including any Firefly outputs, to directly or indirectly create, train, test or otherwise improve any machine learning algorithms or artificial intelligence systems, including any architectures, models or weights."

I'm fairly new to this whole AI art generation thing, but how are we supposed to go about not violating this rule? Are we supposed to not post any work anywhere, whatsoever? 

I can now (as it's not in the Beta) use the work for commercial purposes, so how do I prevent whoever I've sold the work to from using it however they want?

Even if I'm not selling the work and wanted to post it other forums, galleries, social media, etc. I can't prevent someone else from downloading the image and training AI with it.

So the only way I can see any of us not violating this rule is to never post or sell any of our work that's used generative fill - anywhere.

What am I missing?

[jane-e · Accepted solution]

@Encartauk wrote on September 14

I am not a lawyer and I do not work for Adobe, so this is my own personal interpretation:

"ML" stands for Machine Learning, which Google says is a program that can find patterns or make decisions from a previously unseen dataset.

The next paragraph goes on to say in part "you must not and must not allow third parties to use any content, data, output or other information received ... to directly or indirectly create, train, test or otherwise improve any machine learning algorithms."

I removed part of that paragraph to focus on the essence of it, so read the entire run-on paragraph again. To me it means do not use Adobe's Generative Fill to train other models, aka stealing the technology. For my part, I'm not even sure how that happens.

Adobe wants you to use GF, and you will only be posting an image. 

(I also removed the errant comma between "to" and "use".)

Once again, this is my personal interpretation and not official word from Adobe. They have an Adobe logo and the words "Adobe Employee" on their avatars.

Apologies for not seeing your post for over a week. Perhaps others will also chime in.

Jane

[Encartauk]

Thanks for the reply. I didn't eloborate on how much I know /don't know about AI and training, but despite being fairly new to it I have some awareness of certain terms/processes, I just haven't used any other more advance software, such as midjourney, stable diffusion, etc.

My understanding is, people can add images into a 'machine learning' and then when typing prompts for the new image, the AI will generate new images using those images as reference. So, one could have the AI look at a load of pics of Iron Man and Super Man, then create new scenes having them battling, or playing pool, or getting married, whatever...

In Adobe's terms mentioned above it states "...output or other information received or derived from any generative AI features, including any Firefly outputs..." I'm reading that to mean the images (or parts of an image) that their Generative AI has produced. 

So, for example, I recently used the Firefly website to create images of various people from text prompts, and then opened them in the generative fill site to modify bits of their hairstyle, clothing, features etc. (I was basically trying to create realistic images of video game characters) but their faces are largely unchanged from what was originally generated. There's a chance that any one of these people's faces could be very close to an actual person's face (rather than a combination of parts from various faces). If someone takes this image and puts it into a ML then even more images could be created using that person's face, I guess? 

That could then create a situation where someone may have an issue where their face is now being used in all sorts of images, even 3d, video, etc. especially if it's being used in nefarious ways.

My understanding is all of Adobe's generative fill use images from their stock images and all of these should be passed for royalty free use, so I guess the original model as accepted their face could be used in such a way?

I just think it's impossible to prevent a third party from using anyone's work for whatever reason once it's out there on the internet, and these terms don't seem to recognise that.

[jane-e]

Not being a lawyer, I can't even try to address your points! Sorry...

Jane

Forum volunteer

[Encartauk]

Fair enough. I don't know where I'm supposed to go to get an answer from Adobe themselves though. I'll have a look around their site more at some point if nothing more comes from this post.