DALL-E 3 will apparently include some sort of watermarking/metadata indicating the item was created by genAI. There's a mention of the Content Authenticity Initiative, but no specifics - OpenAI is not a member.

(paywall)
https://www.washingtonpost.com/technology/2023/09/20/openai-dall-e-image-generator/

"The release comes amid challenges for the San Francisco start-up, as competitive pressure builds. Traffic to and monthly users of both DALL-E and OpenAI�s flagship chatbot have slowed, as Google rushes a fleet of AI-driven products to users. But by integrating its novel image generator into ChatGPT, OpenAI is expanding its market and offering the technology as a feature to turbocharge its chatbot, rather than presenting the tool as a stand-alone product.
. . .
"DALL-E 3′s improvements make it more difficult for a layperson to identify real photos...�You�re not going to be able to trust your eyes,� said University of California at Berkeley Professor Hany Farid, who specializes in digital forensics and works with Adobe on its Content Authenticity Initiative. But Farid emphasized that the DALL-E 3′s improvements are not cause for alarm because AI gets better at mimicking the real world every six months or so.
. . .
"As part of a voluntary White House pledge in June, OpenAI agreed to develop and deploy mechanisms to identify when visual or audio content is AI-generated, using methods such as watermarking an image or encoding provenance data to indicate the service or model that created the content. DALL-E 3 is experimenting with a classifier that looks at where an image came from or the content�s �provenance,� said Ramesh, a method mentioned in the White House commitments.

These types of mechanisms help identify deepfakes but also can help artists track whether their work was used without consent or compensation to train models, said Margaret Mitchell, a research scientist at Hugging Face and former co-lead of ethical AI at Google."

https://www.theverge.com/2023/9/20/23881241/openai-dalle-third-version-generative-ai

https://techcrunch.com/2023/09/20/openai-unveils-dall-e-3-allows-artists-to-opt-out-of-training/

"Beyond this, DALL-E 3 has new mechanisms to reduce algorithmic bias and improve safety � or so OpenAI says. For example, DALL-E 3 will reject requests that ask for an image in the style of living artists or portray public figures. And artists can now opt out of having certain � or all of � their artwork used to train future generations of OpenAI text-to-image models. (OpenAI, along with some of its rivals, is facing a lawsuit for allegedly using artists� copyrighted work to train its generative AI image models.)"

https://www.wired.com/story/dall-e-3-open-ai-chat-gpt/

https://arstechnica.com/information-technology/2023/09/openai-announces-dall-e-3-a-next-gen-ai-image-generator-based-on-chatgpt/

"Right now, US copyright policy says that purely AI-generated artwork cannot receive copyright protection, so technically any image created with DALL-E 3 will fall within the public domain. While OpenAI doesn't acknowledge that explicitly, it does say that "the images you create with DALL-E 3 are yours to use and you don't need our permission to reprint, sell or merchandise them." That's a marked change from last year when OpenAI restricted DALLE-2 image use based on a license that said OpenAI "owns all generations."

"OpenAI has given no word about its tool's potential to bend the historical record with convincing fabrications, although it says it is experimenting with a "provenance classifier" tool that can help identify whether or not an image was generated by DALL-E 3."

The Content Authenticity Initiative doesn't try to detect AI images but looks for information embedded by its member companies in works created with their tools.

You don't see Midjourney on the list

https://contentauthenticity.org/our-members

The beta Verify tool is looking for tags - and in the case of Photoshop created images which use generative fill, they're tagged and the Verify tool finds that tag.

What I think this means for stock contributors is that we can't use Photoshop's generative fill if we upload work to many/all agencies. Doing special versions for different agencies makes no sense.

Shutterstock and iStock (Getty) are both members of CAI and both forbid uploading AI work. Getty provided an explicit notice on this late in the week:

"As announced in September 2022, Getty Images does not accept files created using AI generative models. This includes Adobe�s recently announced Creative Cloud tools, which are now available with its Firefly-powered generative AI tools built in."

Commercial use, yes. Uploading as stock? See Mat Hayward's answer to my question on this - which was an excerpt from the updated rules for genAI submissions:

https://www.microstockgroup.com/fotolia-com/announcing-adobe-firefly-a-new-family-of-creative-generative-ai-models/msg592314/#msg592314

Shutterstock doesn't accept AI generated images (last update to this page is apparently 8 Sep 2023)

https://support.submit.shutterstock.com/s/article/Shutterstock-ai-and-Computer-Vision-Contributor-FAQ?language=en_US

They may need to say something specific about using generative fill in Photoshop for part of an image and whether they would consider that a no-no

Shutterstock is a member of the Content Authenticity Initiative, so they would likely look at images in CAI's verify tool to see its status

https://contentauthenticity.org/our-members

https://verify.contentauthenticity.org/inspect

Here's what Verify would say about a JPEG file where Photoshop's generative fill was used (this is a test file of mine):

"CONTENT SUMMARY
This image combines multiple pieces of content. At least one was generated with an AI tool.
AI MODEL USED
Adobe Firefly
PRODUCED WITH
Adobe Photoshop 25.0.0"

Getty provided an explicit notice for contributors about use of generative fill in Photoshop this week:

"As announced in September 2022, Getty Images does not accept files created using AI generative models. This includes Adobe�s recently announced Creative Cloud tools, which are now available with its Firefly-powered generative AI tools built in."

With respect to Adobe Stock's rules as Mat outlined them, I'm still not sure it's "safe" to upload something where generative fill was used on the background and NOT mark it as AI generated. I'd be worried about getting my portfolio blocked by the current crazy process where they "shoot first and ask questions afterwards"

My examples and questions in this thread didn't get an answer (possibly because the powers that be don't know)

https://www.microstockgroup.com/fotolia-com/announcing-adobe-firefly-a-new-family-of-creative-generative-ai-models/msg592350/#msg592350

For a stock contributor to upload to multiple sites, effectively the new Photoshop generative tools can't be used. It makes no sense to do multiple versions, and as at least two of the major sites have forbidden their use, what would be the point?

...If I understand it right it seems that the low earnings are due to two main factors, one is that template creators are now a priority and they receive more money and the second part is that it doesn't matter whether you sell more, if others sell more too, you receive less.

I'm not with Canva any more, but the explanations you quoted highlight just what a Canva-favorable contributor-hostile system they have created.

Their system allows them to control their royalty costs so that Canva gets a predictable (and I assume large) share of whatever subscription revenue comes in and avoid heavy customer usage of Canva's services causing a drop in earnings - for Canva. With somewhere like Shutterstock, the more a subscriber downloads, the less profit SS earns. SS contributors earn more as their download count goes up.

With Canva, contributors don't see a linear relationship between the number of downloads/uses of their work and their earnings. Short of banishing all the other contributors to Mars so one contributor sweeps the whole pool, you can't grow your income predictably by growing your downloads. You're in competition with other contributors and the boosted earnings arrangement at any given time.

In addition to the overall number of downloads going up in a given month's pool, Canva can change the weighting for any type of content at any time and that can reduce your payments even if your work is heavily downloaded - infographics are no longer getting a boost; certain areas of the world may not be getting a boost and so on.

I'm sure Canva's investors love this scheme.

If Canva had used this payout model when starting out, they'd have had no content. 123rf did start out with this setup (a pool scheme with no minimum, but no boosted categories). It very quickly changed to include a minimum payout on subscription downloads because the royalties were so much lower than SS's and contributors weren't happy.

...There's no transparency at all and the bonus is a pittance when you take into account we've basically handed over copyright our images to adobe's AI.

There's a lot I don't agree with about how Adobe went about using Adobe Stock images for training data - as I noted before, I think a good lawyer could demolish their argument that the contributor agreement gave our consent to this use - but I strongly disagree that we have effectively handed over copyright to our images.

It's true that if genAI got better it could potentially put us out of business licensing photos/illustrations, but using our images to learn about how images are constructed and what objects look like and how they connect/interact is not the same thing as grabbing copyright.

If any of the commercial getAI tools produced a replica of a copyrighted image, I think there'd be lawsuits. There have already been lawsuits where someone created a very, very similar image.

Without our images, collectively, there'd be no generative AI. Zero. But we still hold the copyright to our work, with whatever value the buyers will place on that over time.

I have an update - my account was unblocked yesterday, which is great, of course.
Unfortunately I got no letter of any kind notifying about it and, most importantly, providing the actual reasons behind the 3-week block, besides one generic response about it being related to AI copyright infringement. I can't seem to find if any images were deleted from the portfolio.

Glad to see that you are freed from the block, but IMO it is absolutely imperative that you get details on (a) what the suspicion was; (b) how many images were affected; (c) what action, if any, was taken. If the answer is that no action was taken, then (d) there should be a sincere apology. Adobe Stock should remember that 3 weeks of earnings could be several hundred dollars - this is bad in all circumstances, but unconscionable if it was all a "mistake".

Given how broken this process appears to be, it really does bear repeating that a better response would be to suspend upload privileges and leave all images in the portfolio live except for the "suspect" ones (which could be rejected temporarily if the software team can't figure out a quick way to render them offline).

This is people's livelihood in many cases. This is a serious issue for contributors. Adobe Stock needs to improve this process, especially for established contributors who've been with them for (let's say) more than one year.

...Examples when to label your image as generative AI:
...-Making significant adjustments or changes to a human subject in an image

This article has two images at the beginning which show Photoshop work with generative fill. Would the changes to the human's clothing shown count as "significant"? I realize the decisions will be subjective, but given the recent account blockings, I think you can understand contributors' nervousness.

https://gizmodo.com/adobe-firefly-ai-photoshop-illustrator-free-ai-art-1850833489

I've done a lot of Photoshop editing and re-coloring over the years. If I continue to do it the "old fashioned" way and submit it without tagging as AI, I'd be worried that I'd get flagged as having used AI but not tagged my content as such. Can that dialog with edit history help here?

I've just updated Photoshop to v25 and will experiment a bit.

The title says Hydroelectric power dam on a river and dark forest in beautiful mountains, Generative AI, but I think some AI design tool came up with this dam and hydro plant - and it doesn't understand the concept.

It's fine to have generic images of a dam and hydroelectric power generation plant, but it has to have a roughly accurate depiction of one. These aren't.









So the above should have been rejected, IMO, but this one works well enough - water flowing over the turbines and down the river:



If it looks like a photo, it can't be at odds with the object it claims to be.

So possibly the earlier examples could be re-titled "Non-working hydroelectric plant stands as monument to wasted money and incompetent leaders" and then they'd be fine?

Just wondering how I could convert Adobe RGB to SRGB? I'm in Adobe Photoshop right now and I'm surprised that this isn't an option when saving. It looks like Adobe RGB is locked in though there is a checkbox next to it.

In the Edit menu, towards the bottom, are two options - Convert to Profile and Assign Profile. They are very different and what you want is Convert to Profile

Good explanation here.

https://arstechnica.com/information-technology/2023/09/microsoft-offers-legal-protection-for-ai-copyright-infringement-challenges/

" 'Specifically, if a third party sues a commercial customer for copyright infringement for using Microsoft�s Copilots or the output they generate, we will defend the customer and pay the amount of any adverse judgments or settlements that result from the lawsuit, as long as the customer used the guardrails and content filters we have built into our products,' writes Microsoft."

Kirsty, I have no idea what triggered Adobe's account blocking, but I took a look at your new uploads on Shutterstock to see if anything stood out given the issues Adobe Stock has been focusing on recently (mostly with AI, but I don't think it's exclusive to genAI uploads).

I wonder if it's the watercolor and other art pieces that reference an artistic style (I realize no specific artist is mentioned). Possibly they overshot the mark with the problems of contributors referencing specific artists and your very general titles - such as Abstract hand painted landscape scene in a minimal impressionist style got caught in the crackdown. This is the sort of thing I'm referring to (obviously can't check on Adobe Stock).

https://www.shutterstock.com/image-vector/abstract-hand-painted-minimal-landscape-japanese-2324511981
https://www.shutterstock.com/image-vector/hand-painted-sunny-landscape-mountains-background-2324510779
https://www.shutterstock.com/image-vector/abstract-hand-painted-watercolour-japanese-themed-2324511257
https://www.shutterstock.com/image-vector/hand-painted-abstract-mountain-landscape-japanese-2318274239

Several of us have suggested that rejecting (retroactively) the supposedly problem content would be better than blocking a whole account, particularly for established contributors. I think the inspection situation has become so crazy in the last few months. As noted above, the good news is others have had their accounts unblocked (although it is taking a while and there's lost income as well as no ability to request a payment in the interim)

Quote from: Wilm on September 07, 2023, 02:02

Unfortunately it doesn't work for me as you describe, but by sampling I now had to find out that I too have "AI generated" images, even though I've never uploaded any.

Wilm try this link to filter your portfolio: https://stock.adobe.com/contributor/201081893/wilm-ihlenfeld?load_type=author&prev_url=detail&filters[gentech]=only&

sorry I can't get this to format properly maybe because of the [] brackets but you can just copy and paste. You only have 3 files marked as Ai so don't be too stressed about it!

You don't need the final ampersand - that's the intro character for a parameter. And if you use escaped characters for the brackets - %5D - then the link will work

https://stock.adobe.com/contributor/201081893/wilm-ihlenfeld?&filters%5Bgentech%5D=only

If you want to show only the non AI content, you can use &filters%5Bgentech%5D=exclude

You can tack other filters on to your portfolio link too - like showing it in download order or newest first:

?&order=nb_downloads

?&order=creation

You only need the question mark once after the URL and then start each additional filter with an ampersand

Quote from: Jo Ann Snover on September 02, 2023, 09:49

For genAI content acceptances, the parade of awfuls continues (just in case anyone was thinking things were improving somehow). Too many images in recent approvals shouldn't have been submitted; given that they were, they should have been rejected (I stopped after page 10; my brain needs a break)
....
Other than being able to boast about the numbers in the genAI collection, I can't see any point, not even for training. Train AI on this content and the results will be even scarier.

Adobe, please rein in this mess. You can be so much better than this if you want to.

I don't think Adobe wants to or has the capabilities to deal with this.

Just recently, they closed their discord channel where contributors could post links to genAI works that shouldn't have ever passed review. At the same time, they implemented a server-wide rule that anyone linking to another contributors' work gets a 24 hour (or longer?) ban

To me, these are clear symbols that Adobe wants to pretend that everything is fine. As long as you don't acknowledge something, it doesn't exist, right?

I'm glad there are still independent forums like MSG where posts like this exist. On Adobe's discord, Jo Ann, you would have been banned.

I am only sporadically on the discord channel but I looked today after reading your post and saw the note with which they closed the QA channel:

"I wanted to let you know that we are closing this channel. Thanks to your help, during the last two months we have got hundreds of great (or awful, depending) examples that we shared with our moderation team, to train them better into recognizing not-so obvious mistakes on Generative AI images. I hope we will see the results of this effort in the upcoming months!
Thanks again for your help!"

I beg to differ with his assessment about the genAI moderation team improving  in the last couple of months.

In the past, this forum had a rule, which I abided by, that you don't post other people's work - they can if they want to get feedback, but otherwise it was a no-no.

My take on the AI content is that it isn't a person's work, and thus doesn't involve the same consideration. Especially given the factory production line approach, it's more manufacturing than creating, IMO.

Possibly we need to talk about what the forum rule should be. I'd like to be able to criticize the inspection process with examples, but that inevitably involves criticizing the contributor too. I'm guessing this hasn't come to a head because it's mostly the new gold-rush contributors whose work has been posted, not people who've been producing stock for a while and know better.

I'm probably done pointing out the Adobe Stock genAI train wreck anyway as they appear to be happy to host logo free rubbish just to get the numbers up.