pancakes

MicrostockGroup Sponsors


Author Topic: Do Adobestock have more stringent moderators or rules on AI submitted content?  (Read 2685 times)

0 Members and 1 Guest are viewing this topic.

« on: January 09, 2024, 07:53 »
+2
I've hit a bit of a snag with my AI-generated content. Lately, about 75% of my stuff is getting the thumbs down, even though I'm using the same tools and upscaler as always. I get that Adobestock is all about top-notch images, and that's 100% the way is SHOULD be. But this sudden drop in acceptance in 2024 feels like a seismic change.  Something's definitely up.

Maybe it's something with DALL-E or Topaz Photo AI, or maybe Adobestock have changed their game. It would be super helpful to get the lowdown on this. Knowing what's going on would let me tweak my process and set my expectations right for what Adobestock is looking for now.


Uncle Pete

  • Great Place by a Great Lake - My Home Port
« Reply #1 on: January 09, 2024, 13:50 »
+2
I've hit a bit of a snag with my AI-generated content. Lately, about 75% of my stuff is getting the thumbs down, even though I'm using the same tools and upscaler as always. I get that Adobestock is all about top-notch images, and that's 100% the way is SHOULD be. But this sudden drop in acceptance in 2024 feels like a seismic change.  Something's definitely up.

Maybe it's something with DALL-E or Topaz Photo AI, or maybe Adobestock have changed their game. It would be super helpful to get the lowdown on this. Knowing what's going on would let me tweak my process and set my expectations right for what Adobestock is looking for now.

You may be correct, that the standards have changed. I have things that I uploaded in recent years, not AI, that are now getting rejections. I can only look and guess that the standards for "grain" have changed and there's some fineness tool that reviewers can now see. What I mean is, somehow they are being shown or there's detection for the many small pixels that AI creates when making the images.

I remember years ago, before AI, when leaves, grass, sandy beaches and water, was often rejected for quality, because whatever software the reviewers had, somehow saw it as grain. Then it seems to stop and things were judged better. Now we have AI images and the same type of rejections for that every-loving "Image Quality" are more often, for me, than they used to be.

AI images are not large. In order to make then big enough, we have to upsize. That means pixel pitch and definition will suffer. I'm sure there are people who can share better answers and how to correct this problem. Since I don't do AI photo quality and hardly anything in the way of fine illustrations, I'll just say, there are people who know how to correct the pixel problems, and get their images accepted at the agencies. AS is the most critical of these quality issues.

My answer? Not very special. Upsize to 4400x4400, using some sort of scaler for resizing and refining, edit, then downsize to 4 or 5MP.

« Reply #2 on: January 09, 2024, 17:21 »
0
I've hit a bit of a snag with my AI-generated content. Lately, about 75% of my stuff is getting the thumbs down, even though I'm using the same tools and upscaler as always. I get that Adobestock is all about top-notch images, and that's 100% the way is SHOULD be. But this sudden drop in acceptance in 2024 feels like a seismic change.  Something's definitely up.

Maybe it's something with DALL-E or Topaz Photo AI, or maybe Adobestock have changed their game. It would be super helpful to get the lowdown on this. Knowing what's going on would let me tweak my process and set my expectations right for what Adobestock is looking for now.

Why not consider that the people who are looking for something are above all the customers? they are the ones who pay... ::)
And many customers don't want AI generated stuff.
Be sure Adobe is starting to know this.

« Reply #3 on: January 09, 2024, 22:26 »
+2
I've hit a bit of a snag with my AI-generated content. Lately, about 75% of my stuff is getting the thumbs down, even though I'm using the same tools and upscaler as always. I get that Adobestock is all about top-notch images, and that's 100% the way is SHOULD be. But this sudden drop in acceptance in 2024 feels like a seismic change.  Something's definitely up.

Maybe it's something with DALL-E or Topaz Photo AI, or maybe Adobestock have changed their game. It would be super helpful to get the lowdown on this. Knowing what's going on would let me tweak my process and set my expectations right for what Adobestock is looking for now.

Why not consider that the people who are looking for something are above all the customers? they are the ones who pay... ::)
And many customers don't want AI generated stuff.
Be sure Adobe is starting to know this.

There is a filter and they can filter non AI stuff if preferred.

« Reply #4 on: January 10, 2024, 08:38 »
0
I am now using stable diffusion for upsizing instead of gigapixel ai. Apparently what I always thought were interesting texture details added by gigapixel are in fact just artifacts. Nice looking, but still artifacts.

Also uploading in much smaller sizes than before.

By the way, I think the customers are using the ai filter to specifically look for ai content. Because the Adobe collection has a lot of very beautiful and creative things.

40% of my time is just research and the Adobe collection looks a lot better than all the other places offering the same old same old.

I hope ai agencies are forced to pay licensing fees, but until then I will get my dues by creating content,  selling and earning  as a revenge for stealing my 10 000 files for training without permission
« Last Edit: January 10, 2024, 09:50 by cobalt »

« Reply #5 on: January 10, 2024, 12:00 »
0
I observe the same as the OP. In my opinion, it's not due to 'technical problems,' which Adobe cites as the reason for rejection. It's because of the sole motive somehow.

I suspect the generative images of the same concept are now repeating for the 100,001st time and are therefore being rejected. Generative images usually don't look very unique (most of the time). However, portraits are still accepted.

 
« Last Edit: January 10, 2024, 12:58 by MadMax »

« Reply #6 on: January 11, 2024, 06:31 »
0
Totally get that but don't you think it would be fair for Adobe to say that instead of the usual "standard" reply?

« Reply #7 on: January 12, 2024, 03:27 »
0
I am now using stable diffusion for upsizing instead of gigapixel ai. Apparently what I always thought were interesting texture details added by gigapixel are in fact just artifacts. Nice looking, but still artifacts.

Also uploading in much smaller sizes than before.

By the way, I think the customers are using the ai filter to specifically look for ai content. Because the Adobe collection has a lot of very beautiful and creative things.

40% of my time is just research and the Adobe collection looks a lot better than all the other places offering the same old same old.

I hope ai agencies are forced to pay licensing fees, but until then I will get my dues by creating content,  selling and earning  as a revenge for stealing my 10 000 files for training without permission

It's not worth worrying. You can't change the current market development.
Even macrostock provider like Getty are doing the same:
https://www.theverge.com/2023/9/25/23884679/getty-ai-generative-image-platform-launch

Getty did a cooperation with Nvidia to develop an own AI image generator, which was trained with Getty's image data base.
The contributors will receive peanuts as compensation.

I can understand it from a market economy perspective.
For the providers, paying less commission (current 15 to 50% of the sales) to the contributors is extremely high saving in the long term.
And look at the first results. They seem to have the potential to outperform Midjourney and Stable Diffusion.
In addition, they prevent the prompting of protected content or famous personalities in order to avoid copyright infringements.

I can even imagine Getty opening a program in the future to hire selected photographers to provide specially needed training material for future model generations.
I mean, why would you still need the mass of contributors in the future?

I'd rather invest more the time in the analysis of market niches or switch to other content media like video right away.

But back to the topic:
Yes, Adobe is much more strict on new high batch volumes of generated AI content.
And the volume of the batches has also an influence on review time or strictness.

Small batches with 1 to 3 images seem to be much faster reviewed (sometimes even in 15 to 30 minutes) im comparison to batches with 100+ files and sometimes are less strict reviewed.
I can imagine that some reviewer are really pissed of seeing batches of thousand of files with all the same stuff. But they probably can't just mark the contributors as spammers.

« Reply #8 on: January 12, 2024, 08:06 »
0
The same here.  A lot of mass rejection of AI generated photos recently. 

« Reply #9 on: January 12, 2024, 08:19 »
0
Declined files that I reprocessed by upsizing with stable and uploaded in smaller size have now been accepted.

@ andrej

The reason. agencies will need content produced by regular contributors with ai is the same they want it for photos - they need an amazing diversity of content and subjects from the entire planet.

You cannot do research on EVERYTHING and just work with paid artists who do work for hire jobs.

It is much easier to have masses of uploaders, pick and choose what you need for different customer groups.

If the "free image" sites did not replace us, neither will ai.

I do think agencies will work with specialized ai producers more closely for elite collections. In the same way they do it with photos now.

There are collections at all kind of price points and licensing. types.

Also...for ai training agencies will keep needing huge amounts of freshly uploaded content, maybe even more than before.



« Reply #10 on: January 12, 2024, 10:05 »
+1
Declined files that I reprocessed by upsizing with stable and uploaded in smaller size have now been accepted.

@ andrej

The reason. agencies will need content produced by regular contributors with ai is the same they want it for photos - they need an amazing diversity of content and subjects from the entire planet.

You cannot do research on EVERYTHING and just work with paid artists who do work for hire jobs.

It is much easier to have masses of uploaders, pick and choose what you need for different customer groups.

If the "free image" sites did not replace us, neither will ai.

I do think agencies will work with specialized ai producers more closely for elite collections. In the same way they do it with photos now.

There are collections at all kind of price points and licensing. types.

Also...for ai training agencies will keep needing huge amounts of freshly uploaded content, maybe even more than before.

You make valid arguments that I would like to respond to.

[1. Photo providers with free images haven't ruined the microstock market yet]

In my opinion, the reason for this is because many users would have used these free, some very professional photos from for example unsplash, but they heard and found out that there are many black sheep, who try to rip-off users through expensive warning letters from lawyers with whom they cooperate.
So they rather invest 2 to 5$ for an image on providers like Adobe to avoid copyright infringements and have quiet nights without insomnia.

[2. Agencies will need content produced by regular contributors]

Absolutly right. But they neither need millions of new images, which are copycats of historic generic bestsellers or content, which obvious won't find any or very few customers. With AI it is extremly easy to generate mass of such copys.

I don't believe they will need every content and subjects on the earth.
They will even more need creative concepts of current trends like diversity, sustainability, environmental awareness, home office work, ethical consumption or new coming global trends. I can't imagine they still need the million and one easter background or close up of a cocktail on a beach bar.

And that is exactly the bridge to:
[3. Agencies will work with specialized ai producers more closely for elite collections]

That's actually roughly what I meant.
First they will try to work more together with skilled photographers, who can produce elite content, which can't be generated by AI.
And second for sure with a small fraction of very skilled AI contributors, who can create very original content with professional image und retouche editing.

[4. It is much easier to have masses of uploaders, pick and choose what you need for different customer groups]
In earlier times, before the mass flood of images, absolutely yes.
But nowadays it's becoming increasingly difficult to pick out exceptional good images from the masses.
That's why it would only make sense to concentrate on contributors (point 3) who provide on usual very demanded content.

[5. Agencies will keep needing huge amounts of freshly uploaded content for ai training]
This is an interesting and valid argument. Probably they will accept in future only content, which can increase the quality of the AI generator.
That would be rather specific motifs.
Or they could resell their storage with unwanted stuff, which haven't been sold for years in the past, to smaller tech companys, who want to develop own AI models and avoid copyright infringements. This could also be an additional source of income for the agencies.

@ Back to topic:
The resolution has also an impact of the acceptance ratio.
I upscale now the AI images first 4x, retouche obvious artefacts or generative errors and scale down by approx 0.6 to max. 0.75. The max. resolution is then 5376 x 3072 pixel.
I believe most time the problem is the upscale noise, generated by Topaz AI. Upscayl seems to generate less noise but the images are sometimes over-sharpened.

« Reply #11 on: January 12, 2024, 10:22 »
+2
@ andrej

I think we agree on many things.

Already the agencies have different levels of prices and collections, sometimes even different agencies under the same umbrella.

Usually one cheap collection/agency where they also get all the newbie content, but that content also has a lot of "authentic real life images" that work exceptionally well if you need content for social media. Files that are too spectacular or studio stuff will break the flow of a social media feed if you want to "connect" with normal people. You need content that blends in.

That was the great advantage of eyeem, they have millions of real world content images, that many photographers will look down on but people in marketing desperately need.

And Getty was selling these "ugly realism" files in a high priced collection.

It will be the same with ai.

There will be also be needs for "amateur looking" ai content, including endless copies. Not just the super high end stuff.

It is IMO an illusion to believe customers only want super high quality studio productions, wether it is photo or ai.

They need a huge variety of content.

And large agencies need a huge inflow to really have their editors dig down into the collections to put together some amazing selections that work brilliantly of each other.

And they need new files all the time.

That is why having so many duplicates of popular content is less of a problem than most people think.

The algos will quickly sort through what comes in, if it is not interesting it will drop down into oblivion quickly.

The algos also present customers with a good mix of new, old and ai. It is the biggest secret of every agency how they create that.

Most agencies also have higher priced elite collections or even separate agencies with high quality content where they also offer more service to the customer with personal editors.


The thing we sell, even more than images themselves, is TIME.

Customers have no time. Otherwise they would use their own smartphone and take a shoot of a strawberry smoothie and not pay hundreds of dollars for a simple image on Getty (happened to me, got over 100 dollars from eyeem for the most simple iphone snapshot).

Everybody has their own view of where the market is going and with the rise of ai, many producers will say - that is it, I am quitting.

It is a similar situation to when microstock came on the scene and all the old timers absolutely haten on us how we destroyed their wonderful and exclusive way of life.

Personally I believe that the opportunities far outweigh the risks coming with the change.

I also don't believe normal stock photography is dead, because when I do my research I see HUGE gaps in literally everything, including generic style content over all agencies.

And there is video, editorial etc...

So, to all of us...happy 2024 and the dawn of a new age in art.

« Reply #12 on: January 12, 2024, 15:38 »
+5
The same here.  A lot of mass rejection of AI generated photos recently.

I think they just put on their "see quality issues everywhere" glasses for all types of content recently. I had a batch of 7 photos rejected yesterday for quality issues - it only took them about 5 days to reject them though :)

The review system is completely unpredictable

« Reply #13 on: January 14, 2024, 07:04 »
0
I had 5 files reviewed in 5 days. 3 accepted 2 declined. Now I can look for the problems in the declined ones before I upload more from the series.

This is a much better system than before.

« Reply #14 on: January 14, 2024, 09:07 »
+1
@ cobalt

As you already have more experience, have you already successfully reuploaded images, which were rejected on the first?
Have you just downscaled them or run denoise, etc. procedures?

I might try uploading smaller batches with differently edited images soon, from which I can learn what is paid more attention to. However, there is probably still the problem of the dispersion of reviewers.

If Adobe had a detailed guide with sample images to learn from, you could build automated actions for mass editing in Photoshop. I mean their current guidelines are just to generic written.

https://helpx.adobe.com/stock/contributor/help/generative-ai-content.html
https://helpx.adobe.com/stock/contributor/help/photography-illustrations.html
https://helpx.adobe.com/stock/contributor/help/the-review-process.html

Interesting is that they seem in theory also evaluate the commercial value, title and keyword accuracy, aesthetic properties, uniqueness, what would be really great, but in practice only technical reasons are mentioned.

« Reply #15 on: January 14, 2024, 10:45 »
+1
Woke up this morning to find out 24/24 recent AI photos are rejected.  I use Gigapixel AI 4x upscaling, but I'll try Midjourney's own upscaling in 2x and see what happens.  I need to find the best way to upsample AI images to avoid mass rejections.

« Reply #16 on: January 14, 2024, 12:01 »
+1
Some may have read my numbers about rejections in the other thread https://www.microstockgroup.com/38186/38186/msg596960/#msg596960

Today (Sunday) the numbers are in my case getting worse: appoved 12, rejected 48 (2 of those rejections have been uploaded only 3 hours ago)

So the complete numbers of rejections for the last 10 weeks in my case are:
05.Nov 2023 to 12.Nov 2023 - accepted 287, rejected 13 - rejection rate: 4.3%
12.Nov 2023 to 19.Nov 2023 - accepted 223, rejected 57 - rejection rate: 19.6%
19.Nov 2023 to 26.Nov 2023 - accepted 180, rejected 32 - rejection rate: 15.1%
26.Nov 2023 to 03.Dez 2023 - accepted 245, rejected 12 - rejection rate: 4.7%
03.Dez 2023 to 10.Dez 2023 - accepted 150, rejected 28 - rejection rate: 15.7%
10.Dez 2023 to 17.Dez 2023 - accepted 175, rejected 21 - rejection rate: 10.7%
17.Dez 2023 to 24.Dez 2023 - accepted 166, rejected 126 - rejection rate: 43.2%
24.Dez 2023 to 31.Dez 2023 - accepted 217, rejected 95 - rejection rate: 29.5%
31.Dez 2023 to 07.Jan 2023 - accepted 252, rejected 249 - rejection rate: 49.7%
07.Jan 2023 to 14.Jan 2023 - accepted 261, rejected 308 - rejection rate: 54.1%

I really wish I could get more than "quality issues" as a reason (that was 99.8% of the reason for rejections)
The nature of my images hasn't changed and the greatly increased rejections since Dec 18th are for the same image sets that had images accepted right before.

I think that the internal guidelines have become a lot stricter - and maybe that's a good thing.
But something more than "quality problems" would be necessary in order to improve ourselves.

I am now also trying smaller scaling of the MidJourney results, more in the x2 range than x4 before.
This should produce fewer artifacts and therefore cleaner images.

Unfortunately, it remains a guess as to whether we were right - unfortunately, we lose a lot of images that could be online until we figured out what the "exact" rejection reason might be.

Have a sunny day,
Michael



« Reply #17 on: January 14, 2024, 12:17 »
0
Personally - I think there are just a lot more people trying to "get rich quick" from "ai", and so with 100x the submissions, they are just randomly rejecting stuff. I don't think they will admit that - but I suspect that is probably the case.

Uncle Pete

  • Great Place by a Great Lake - My Home Port
« Reply #18 on: January 14, 2024, 15:18 »
+2
Personally - I think there are just a lot more people trying to "get rich quick" from "ai", and so with 100x the submissions, they are just randomly rejecting stuff. I don't think they will admit that - but I suspect that is probably the case.

If it isn't just random, maybe they have higher standards, because there is so much being uploaded? They can pick and choose. Any noise or grain, artifacts can be a problem.

Also I consider that the problem is upscaling, no matter what or how, these are small to start with, why do people think they have to be made, extra large? Why not upload 4MP images, for example? If that what it takes to get them to pass.

« Reply #19 on: January 14, 2024, 18:13 »
0
It used to be that larger images have better sales, because customers can sort by size. I always try to at least have 12mp/XL image.

But since the upsizing skills of photoshop have also improved, maybe uploading ai with 4mp is enough.

@andrej

Yes I have had reprocessed files accepted. Upsized with stable instead of gigapixel and uploaded in a smaller size.

Also most of my more recent uploads have been accepted. Except today, 2 of 7 declined. But both illustrations went through, as have the other illustrations before them, so at least I seem to have a workflow for illustrations. The files declined where the visually weaker files, I don't think the decline was artifact related. I have others from the series, maybe they work better.

But the fast turn around makes all the difference. Give the reviewers a first test batch, then gradually and very carefully process the rest.

We will all benefit from higher quality standards.

I just have to accept that thousands of files prompted last year cannot be saved, the quality is now too low.

But if more agencies start taking ai, then maybe there will be some places that have less strict quality demands for cheaper prices.

At the moment Adobe is the only real place to make money with ai.

wds

« Reply #20 on: January 14, 2024, 21:03 »
+3
I think the other agencies will start to take AI submissions when they see the additional revenue that is generated. They can't expect customers to sit and "waste their time" by trying to generate their own AI as opposed to having a plethora of ready to go AI images to choose from.
« Last Edit: January 15, 2024, 11:32 by wds »

« Reply #21 on: January 14, 2024, 21:48 »
+1
am sure Getty/istock will take ai once their legal team has found a way to make it work for them.

If the nvidia generator is good, they might restrict themselves to content created from it.

But if they want to have a good collection, they will need enough suppliers sending in good content.

Shutterstock, at least according to their last comments on their financial reports, seems to think just licensing their media bank for training ai is their new business.

They made 80 million selling stock but already an additional 20 million licensing data. And they seem to be sharing just a minuscule amount of that with creators.

I personally think that is a dead end. Only allowing customers to create content will not build up a stellar ai collection.

But whoever had a great business brain seems to have left for now. They still have editorial and video/pond5 though. And with new and good managers they could turn that ship around very quickly.

The smaller agencies , deposit, envato etc.. will probably start taking ai at some. Perhaps like dreamstime without people content.

At the moment Adobe reigns supreme. They made the best decision and have increased the value of their Adobe stock offering immensely.

And of course the tight integration with ai photoshop tools.


« Reply #22 on: January 15, 2024, 00:24 »
+1
am sure Getty/istock will take ai once their legal team has found a way to make it work for them.

If the nvidia generator is good, they might restrict themselves to content created from it.

But if they want to have a good collection, they will need enough suppliers sending in good content.

Shutterstock, at least according to their last comments on their financial reports, seems to think just licensing their media bank for training ai is their new business.

They made 80 million selling stock but already an additional 20 million licensing data. And they seem to be sharing just a minuscule amount of that with creators.

I personally think that is a dead end. Only allowing customers to create content will not build up a stellar ai collection.

But whoever had a great business brain seems to have left for now. They still have editorial and video/pond5 though. And with new and good managers they could turn that ship around very quickly.

The smaller agencies , deposit, envato etc.. will probably start taking ai at some. Perhaps like dreamstime without people content.

At the moment Adobe reigns supreme. They made the best decision and have increased the value of their Adobe stock offering immensely.

And of course the tight integration with ai photoshop tools.

After releasing Midjourney V6 they already lag behind on every level of AI generated image.

« Reply #23 on: January 15, 2024, 03:05 »
+1
@
I wouldn't bet on it in the long term.
Nvidia is a heavyweight cooperation partner that brings a lot of technical and theoretical know-how in the software/hardware industry and model development of AI.
NVIDIA Picasso for example seems also to generate 3D models and has a web integration of user applications like photoshop, etc.
https://www.nvidia.com/de-de/gpu-cloud/picasso/

If they manage to generate the 3D objects photorealistically, I see even more potential here, because you can generate entire scenes in which you can adjust every detail as desired (think of product visualizations, for example, in which you can rotate the objects or move the light source as desired).

//Edit:
It is interesting to note that Cuebric has been developed with Nvidia for products for film production.
Looks extremely promising there.
https://cuebric.com/disguise
« Last Edit: January 15, 2024, 03:21 by Andrej.S. »

« Reply #24 on: January 15, 2024, 04:22 »
0
I think nvidia might be the company to give us the best ai video engine longterm.

Getty was wise to strike up a cooperation with them.


 

Related Topics

  Subject / Started by Replies Last post
5 Replies
6721 Views
Last post March 23, 2006, 16:02
by admart01
3 Replies
3765 Views
Last post February 24, 2010, 07:15
by macrosaur
28 Replies
10830 Views
Last post May 01, 2012, 12:46
by lagereek
7 Replies
6063 Views
Last post October 10, 2019, 21:32
by dragonblade
7 Replies
2733 Views
Last post December 01, 2023, 14:05
by Evaristo tenscadisto

Sponsors

Mega Bundle of 5,900+ Professional Lightroom Presets

Microstock Poll Results

Sponsors