It might seem fruitless to urge new rules for genAI content when almost none of the current ones are being followed or enforced, but here goes.

Some locations - oil refineries, factories, research labs, outer space - are hard to access for stock photographs. That makes them ripe targets for the genAI factory producers who churn out content based on copying someone else's title and making it a prompt. I know Adobe says "don't do that" but there's a lot in the collection already.

I took a look at some examples of robotic arms in a solar panel factory - or what purported to be that. I then realized I don't know anything about the details of solar panel manufacturing, but did a little searching online to confirm a gut feeling that the genAI copycat content was rubbish. It looks high-tech-ish and robot-ish but it isn't real and arguably would harm the credibility of any buyer who licensed it to use with an article about increasing use of solar panels.

While looking at the human-produced solar panel factory images on Adobe Stock I recognized some of the prompts as ones used for genAI images. I took two and did searches and made screen shots to give a visual example of what I'm talking about.

It's possible this content would be OK if Adobe put a visual label on all genAI content in search results - to allow anyone who needs accurate images to avoid these. It's possible it should go on the no-no list - like specifying specific cities or famous places. The temptation is significant because of the lack of supply of the real thing, but I'm not sure that is enough to make this type of fake stuff OK to offer to buyers. And I'm not a fan of leaving it up to the buyers - how on earth are they supposed to separate the snazzy looking image with the copied title from the real thing?

I think stock agencies accepting genAI content need to think hard about setting buyer-friendly, trustworthy, sustainable policies about these sorts of issues

It is especially galling that the sort orders of "Relevance" and "Featured" put some of the newer genAI items ahead of the real images of solar panel factories.

The copied titles from the original (human-generated) content:

Wide Shot of Solar Panel Production Line with Robot Arms at Modern Bright Factory. Solar Panels are being Assembled on Conveyor.

Large Production Line with Industrial Robot Arms at Modern Bright Factory. Solar Panels are being Assembled on Conveyor. Automated Manufacturing Facility


Click for larger version - first one and first four images, respectively, are the human created photos.




Note: I can't be certain all the genAI images are all wrong, but I looked at a bunch of images online and accompanying articles about solar production and did not see anything that looked like what Midjourney (or whoever) came up with. Given that reviewers can't be expected to know the innards of a whole variety of factories or industrial processes either, I'd argue that points towards disallowing this type of content altogether

Some locations - oil refineries, factories, research labs, outer space - are hard to access for stock photographs.

Outer space is another area, where you probably won't get very authentic images with AI, but where I could imagine a large demand for fantasy images, for example impressive space nebulae, or planets with two moons and stuff like that for science fiction stories.

I wouldn't ban the content, just the description of it. So all the thousands of genAI images that say they're Paris, London, Seattle, Rome, Persepolis, etc. could still be uploaded but with general descriptions, not country name, town name, monument name, etc.

So if you searched for Big Ben, London, you wouldn't get any AI images at all. And I don't think images are well labeled on Adobe Stock - when you look at a page of search results you have no idea that there's any genAI content in there. If you just search, AI is on by default - you have to know to turn it off. And there's nothing - unlike with Editorial, or Premium - that shows you which images in a page of results are AI.

I am aware that having stock models pose as scientists in a lab isn't accurate either, but the sale of the fakery with AI is so vast - the entire factory, all the robots, the solar panels, etc. - and the surface appearance of reality is so different that I don't think it's a relevant comparison.

The example images should be labeled generically - "Modern factory with robot-controlled assembly lines" would have been fine

A better labelling of what is ai might be useful, I agree.

But if they don�t do it for editorial, then maybe the customers don�t need it.

I am more concerned about all that stuff with logos and clear copyright violations coming through.

Also photo ai that is labelled as a specific place.

Artistic illustrations of the Eiffel tower in creative variations are fine with me. Art is not a photo.

But you cannot call a photo ai Big Ben, because it is simply not a photo of the real thing.

This image for instance is very beautiful and authentic looking, but it is an ai image and not real.

The Brandenburg gate has 6 columns.

It should perhaps be labelled " headshot of woman in front of a german style monument" or something similar.

https://stock.adobe.com/de/images/headshot-portrait-photography-of-a-pleased-woman-in-her-50s-that-is-with-the-family-in-front-of-the-brandenburg-gate-in-berlin-germany-generative-ai/610015374?prev_url=detail


On the other hand...would keywords like Berlin, Germany even Brandenburg gate be fine? Because otherwise the customer cannot find it. At least Berlin should be includable.

Same for images with the Eiffel tower - The description could be "French style radio tower with interesting metal structure", but the keywords would need Paris, France at least so a customer can find a creative photo ai version of it.

Maybe Adobe could make another blog post specifically for prompting with real places or how to best describe and keyword it.


eta

here is the search for brandenburg gate woman only ai

https://stock.adobe.com/de/search?filters%5Bcontent_type%3Aphoto%5D=1&filters%5Bcontent_type%3Aillustration%5D=1&filters%5Bcontent_type%3Azip_vector%5D=1&filters%5Bcontent_type%3Avideo%5D=1&filters%5Bcontent_type%3Atemplate%5D=1&filters%5Bcontent_type%3A3d%5D=1&filters%5Bcontent_type%3Aaudio%5D=0&filters%5Binclude_stock_enterprise%5D=0&filters%5Bis_editorial%5D=0&filters%5Bfree_collection%5D=0&filters%5Bcontent_type%3Aimage%5D=1&filters%5Bgentech%5D=only&k=berlin+brandenburg+gate+woman&order=relevance&safe_search=1&search_type=filter-select&get_facets=1

In many images the brandenburg gate is wrong, or the quadriga on top etc...


eta

and then there is this image in the search which has nothing to do with berlin, but has berlin in the keywords

https://stock.adobe.com/de/images/bride-and-groom-standing-in-front-of-a-heart-shaped-cake-generative-ai/570122192?prev_url=detail

But they are buying ai images.

Apparently in record volume, hence why you have so many youtubers screaming about it and showing their amazing results.

ai is also the most important art trend this year (this century?), so it is very high demand content.

. . .

In many images the brandenburg gate is wrong, or the quadriga on top etc...


eta

and then there is this image in the search which has nothing to do with berlin, but has berlin in the keywords

https://stock.adobe.com/de/images/bride-and-groom-standing-in-front-of-a-heart-shaped-cake-generative-ai/570122192?prev_url=detail

� propos Brandenburger Tor...



Count the columns to believe - I shot it myself 

... just the quarreling couple is Midjourney's

Quote from: Her Ugliness on September 01, 2023, 04:22

Quote from: stoker2014 on September 01, 2023, 03:21

That's why buyers won't buy a lot of AI-generated images.

But they do.

https://petapixel.com/2023/06/06/ai-images-are-outperforming-photos-on-adobe-stock/

Possibly.

"All the above data has to be taken with a pinch of salt. It is compiled from Stock Performer�s customers who do not represent all stock contributors and not all Stock Performer customers choose to hand over their performance data."

 Yes, maybe for whatever reason costomers only buy a lot of AI images from Stock performer contributors and not from anyone who is not a Stock Perfomer's customers.  But I somehow doubt that.
It's also in accord with my experience. My portfolio is about 90% real photos and 10% AI images. Yet by now I sell slightly more AI images than real photos each day, so customers buy a very unproportional high amount of AI images from me. (And no, it's certainly not because my real photos were of poor quality. I do microstock full time, so you can imagine that in order to live from my microstock income I would need to be able to produce images of at least decent quality)

So from my observation customers buy a lot of AI images. Like, really  A LOT.

I understand very well that the ones of us who are aware that AI will eventuelly be the downfall of microstock and therefore the downfall of all of us who do microstock full time really hoped that AI images would not be well accepted by customers and they would prefer real photography, but sadly it's really not what seems to be happening right now. I know there are customers who refuse to work with AI images. I even know a very big world-wide company cancelled all their associations with Adobe because of the AI image mess. But most customers seem to be crazy about AI images and buy them like hot cake.

If you look at Adobe Stock's genAI content sorted by downloads, it seems (to me) to show that there's a huge gap between the bulk of genAI acceptances and what's in the top sellers. I can see why those top downloads would sell - they're useful stock and do not scream "I was created by genAI".

Well,  AI images have only been on Adobe for about a year, while real photos have been around for 19 years (Fotolia time included), so I do not think you can go by the download numbers (yet), as the real photos have a great lead on AI images. My real photo bestseller on Adobe has sold 1003 times. My AI bestseller has only sold for 99 times. But my real photo bestselles had the chance to collect downloads for a couple of years, the AI image only for 6 months.
And then there is also a huge over-suppy of AI images so individual AI images will have a hard time collecting so many downloads as there is too much new competition added at the same time. But it doesn't mean that overall customers weren't buying a lot of AI images.

@gameover

this is the smart way to do localized ai images. Use a real image as a background and just ai content at the front.

Great work!

Thank you for the flowers as we say...happy to oblige...

Quote from: Jo Ann Snover on September 01, 2023, 10:50

If you look at Adobe Stock's genAI content sorted by downloads, it seems (to me) to show that there's a huge gap between the bulk of genAI acceptances and what's in the top sellers. I can see why those top downloads would sell - they're useful stock and do not scream "I was created by genAI".

Well,  AI images have only been on Adobe for about a year, while real photos have been around for 19 years (Fotolia time included), so I do not think you can go by the download numbers (yet), as the real photos have a great lead on AI images. ...

I'm not comparing genAI downloads to other types of content, only looking at what is selling best from the content that is marked as AI.

As such (and I don't know any numbers, just relative position in a sort by "Downloads"), I think it's a reasonable comparison of which types of AI content are most appealing to buyers.

In other words, everything in the list I'm referring to is December 2022 or newer.

If anyone who has an item in the top 100 would care to share their download number for that image, we'd have an idea of magnitude for AI sales. That's the only way anyone outside of Adobe can know anything about absolute download numbers as they aren't visible to the public.

similarly, many ordinary microscope images are of stained-prepared objects

I still see that as a representation of the real world. It's an example of real world staining - something that many scientists and hobbyists do in labs etc every day. So I don't see that as misleading. Someone may be looking for an example of a specific type of staining. And of course there are many types of microscopy such as bright field, dark field and polarisation that will give very different 'looks.' And I would consider each of those 'looks' to be authentic. However, I would label them appropriately in my descriptions if I submitted some imagery as stock - ie dark field imaging.

...sorry mistake...