MicrostockGroup Sponsors


Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.


Messages - MPfoto71

Pages: 1 2 [3] 4 5 6 7 8 ... 21
51
Yes, but can photos after editing with DxO PhotoLab or Pure Raw be checked with any tools to see if it leaves any traces after using the Ai tool? Same with Topaz Ai.
You can recognise from the image itself (with a certain degree of certainty) whether it was a generative AI that created the image or not. There are algorithms that search for typical artefacts in the image that result from the use of generative AI. This is therefore technically possible even without looking at Exif data or embedded watermarks - only the agency itself knows whether an agency actively uses this.
The trace is therefore the image itself or the specific arrangement of certain structures.
And yes, in order to recognise this via the pixels, an AI is used that is trained to distinguish between a photo and an AI image - we have probably all experienced this before, where a real photo was rejected because it was supposedly taken with AI. In these cases, the recognition AI was simply wrong :)

Yes, but Sora creates images from scratch from other images and there is a greater risk here. I'm more interested in the safety of photos and videos improved using Ai tools.
Yes, that's right and something very similar happens when you denoise with generative AI.
Let me try to describe the process (very simplified) for denoising:
-> the process takes a block of 32x32 pixels with image noise from the original and generates a block as similar as possible without image noise and adds it there, then it continues with the next block in the original image, etc.

The resulting image is therefore completely newly generated and may well contain deviations from the original (see example additional stars in the sky) - regardless of whether we denoise, sharpen or scale with these tools, the technology is always exactly the same, only the training data differs.

52
So you wouldn't even trust RAW file denoising programs because of their Ai tools? Like DxO PhotoLab or Pure Raw. They are the leaders and they boast that they have databases of billions of photos.
If you ask me privately, I generally only believe what I can verify - but as a mathematician, that's my own personal quirk ^^ Basically, I see every generative AI (trained with billions of images) as a potential legal problem in the future, but I'm also very cautious.

The results of these generative AI algorithms are impressive and certainly usable, in some cases (I would not denoise a night sky with stars with a generative AI, then I have some new stars in the sky afterwards and with luck also a few extra comets ;-)

At the moment, everything is unclear from a legal point of view (for example, it is not at all clear where Sora got all the videos for their training) - committing to something here is currently on shaky ground.

53
Yes, I know it's worth being skeptical. Okay, but what will it change if, years later, it turns out that a well-known company that uses Ai tools for image scaling and enhancement was illegally using a photo database?

Would we authors also be complicit with stock agencies? After all, we would no longer remove improved photos or videos using Ai tools from our portfolio, which sell to customers, etc. By purchasing the Ai utility program, we should be released from liability. That's how I understand it. What is your opinion?
It is currently very difficult or even impossible to answer this question. On the one hand, there are currently numerous lawsuits in court around the world concerning the use of images for AI training, some of which are being brought by artists, photographers, graphic designers, etc., because the big companies simply download the entire Internet using web crawlers and use it as training data - but the problem has already been discussed extensively here.

Then there is the problem of the different regions, for example the EU has much stricter regulations than the USA, so does the company from the USA have to comply with the stricter rules for the EU customer of the software or not? Good question...

What happens now, if it turns out in the court cases (which I don't believe) that the companies are really ordered to use only copyright-free images in training and to prove this, would basically all images processed with AI tools (which were trained with unauthorised images/videos) not be legally flawless in the first place. In this case, the software customer would then have to claim the corresponding costs from the software provider (who used unauthorised data) - what a mess that would be, even if it were the right way :)

And to avoid such problems with my software, I don't use such tools, even if you can get some of them for free on the net and integrate them into commercial products.
Just because you can do it and it's possible doesn't mean you should do it :)

54
So why wouldn't Topaz Ai also legally license the photos to Ai. This company is big and has a million users. A lot of people are making money from these Ai tools, but istock is still not clear about these Ai tools.

I will be happy to test these Ashampoo programs, they are also interesting.
The difference is in the way the AIs are used, but of course you can't see that from the outside.

Self-learned generative AI models require vast amounts of images to achieve good results with such a system, and by that I mean ranges of several hundred million images as a lower limit.

The AI system used in Zoom #2 pro is a non-generative system, which takes the information exclusively from the current image to be scaled (+ training images that can be specified additionally). This means that no large training volumes are required, but no new details are generated into the images.

Unfortunately, the topic is very complex and you need a lot of basic knowledge if you want to make well-founded statements here.

I did not want to claim that the various providers have not all acquired the training images legally, but you should at least remain sceptical.
The large image datasets that can be acquired from the relevant companies with 4-6 billion images cost 2-3 digit million sums - only the really big ones can afford such datasets...

Actually, I just wanted to say: Remain sceptical - but I probably didn't manage that very well ;-)


55
Those of your recommended scaling programs also ashampoo zoom and zoom pro have Ai.
That is correct, but the training data used in Zoom #2 Pro and Ashampoo Zoom #2 are only self made images.

How do I know that?
How do I best express this... I developed the Zoom #2 programme, so I can say with certainty that no data with third-party copyright was used.

56
Using an AI tool for sharpening certainly does not alter a picture in a way that you lose copyright, strictly speaking or otherwise.

That no pixel of the original image may be retained is hardly an argument, because that if true of a plethora of filters or other modification.

If you apply a filter to a photo, so that it looks like an impressionist painting with Photoshop, no pixel is the same either. Even if you just make a picture darker or brighter or change the colour temperature, no pixel may be exactly the same afterwards, if you change it enough. But you still have the copyright.
The rights problem (which has not yet been finally clarified) also arises from the question of where the training data comes from. As soon as an image has been processed with a generative AI algorithm and not all of the image rights (without exception) of the training data are held by the creator of the software, this is a potential problem.

If a purely mathematical algorithm is used, no training data is required and the question of rights does not even arise.

I don't want to argue against you here, I just want to sensitise users to the problem. As I create image editing software myself (also with AI algorithms) and sell it, I deal with this problem almost every day, always ensuring that the rights issue is fully addressed for my users :)

57
Maybe strip the meta or exif data before submitting.
Removing the corresponding metadata (Exif tag Software used) could work.
Here is a screenshot of our own Zoom software where you can see that Topaz leaves an entry that is possibly recognised by IStock and is considered AI-generated.

In fact, this is technically correct, because the AI system is very similar to a Ki image generator. In the end, not a single original pixel of your photo is retained - the image is generated block by block in a different resolution/de-noised/business.

Strictly speaking, it is no longer your image, but a completely generated image to which you no longer have the rights.

If you want to avoid such problems, you would have to avoid scaling/de-noising/sharpening programmes with generative AI and rely on other algorithms.

Examples of such scaling programmes with non-generative AIs would be Zoom #2 professional or Photozoom #8 and in the area of denoising, Neat Image or Denoise #5 professional come to mind.

58
DepositPhotos / Re: Does anyone get money for AI?
« on: December 28, 2024, 17:43 »
Yes, that's right - I assume that the dataset was earnings, I had no luck with Deposit's support, they didn't express themselves very precisely, but didn't deny it either...

59
DepositPhotos / Re: Does anyone get money for AI?
« on: December 28, 2024, 15:52 »
@Stoker2014:
I had a dataset earning at Deposit as "kundenspezifischer Verkauf" (sorry for the german text here) on August, 2nd 2024.

Maybe you want to check your earnings stats too, it should be there...

60
123RF / AI/ML Earnings reduced in stealth
« on: December 26, 2024, 07:06 »
Hello everyone,

Today (26.12.2024) the AI/ML income booked on 19.12.2024 was reduced from $79.66 to $15.26 on my 123rf account.
(until yesterday 25.12.2024 it still read $79.66, so suddenly $64,40 are missing)

Have a look and see if something was taken away from you for Christmas... I would say that they are currently correcting the annual figures and needed something on the balance sheet.

I have already written to 123rf support, but I have no hope that this will do any good. (I've been waiting for image reviews since February 24 and nothing has happened)

Merry Christmas :)

61
Everything looks "kinda" normal for me on the map of the last 20 sales from yesterday and today - 60% Europe / 20% US/NA / 20% Asi

62
Adobe Stock / Re: What's your weekly ranking and how many images?
« on: December 18, 2024, 17:33 »
@Cobalt: You are very far ahead - very nice! :)

63
iStockPhoto.com / Re: November stats are up
« on: December 18, 2024, 15:53 »
Just to keep you informed - the support answered my ticket with serveral specific questions:

"Hello Michael,
For more information regarding our generative AI services we have created a new FAQ section to address the questions you might have. 
Please visit the link below: https://contributors.gettyimages.com/article/10953
Best Regards."


And my ticket was closed wihout any answer to my questions...

64
iStockPhoto.com / Re: November stats are up
« on: December 18, 2024, 11:41 »
Perhaps we are misunderstanding the 50%.
I have an ai sale for 6 dollars, I got 1.18
Or maybe is was several ai creations.
50% of 6 dollars is 3 dollars.
This gets equally split over all involved files. If 3 files were involved we each get around 1 dollar.
Maybe it works that way?
These are all possibilities, but unfortunately we don't know what they are - the 50% figure is in the air, but if I'm seeing things correctly, no one is even close to reaching it.

I've already written to the support team and I think everyone should do that so that they realise that something is wrong here :)
It's highly unlikely that this will have any effect, because we're being ripped off anyway.

If, as in my case, this $11.38 is supposed to be the total amount for 1 year of Dataset Earnings, it's not even close to being okay.

65
iStockPhoto.com / Re: November stats are up
« on: December 18, 2024, 08:34 »
@Stoker2014: Thanks for the hint with the .pdf

I found these two lines there:
- AISERVICES United States of America 41.07 varies 6.54
- AISERVICES varies 30.37 varies 4.84

In both cases ($6.54 of $41.07 and $4.84 of $30.37) it is 15.93% of the total - so neither 15% (photo) nor 20% (video)

And the message from Getty/IStock says Training share: 50% of the revenue is split across all files used to train our AI services, with each file receiving an equal share.

Something is very wrong here!

66
iStockPhoto.com / Re: November stats are up
« on: December 18, 2024, 07:45 »
I just ran my stats through TodayIs20 (thanks @ShadySue for the tip):

The following entry can be found there:
- AISERVICES 11.39$ 6 sales

I have roughly the same portfolio on all platforms - about 25,000 photos and 4,500 videos (without AI material) - these have generated Dataset Earnings in 2024 as follows:
- Pond5: $1,141.42
- AdobeStock: $526.98
- Shutter: $292.20
- BigStock: $114.77
- Deposit: $114.25
- Alamy: $100.82
- Getty/IStock: $11.39

Getty/IStock's payout to Dataset Earnings for October 1, 2023 to September 30, 2024 (according to Getty) is truly the biggest joke of the year.

An extra Exiting News mail would actually have been due for such a rip-off of the contributors!
I'm really lost for words here - and that doesn't happen often :)

67
Adobe Stock / Re: AS Contributor Survey
« on: December 16, 2024, 08:06 »
@Raul:
My wife and I both have our own AS account and neither got the email about this survey (not even in the spam folder), the communication in the account should be active.

Through the link (thanks for that) we were both able to participate.

I'll join Jo Ann here and also ask if we can know what the aim of the survey is :)

68
Shutterstock does what's good for Shutterstock - they don't care about the contributors, I think we can all agree on that :-)

I interpret the resubmission of already rejected content as follows:
-> I am not allowed to submit the exact same file a second time because it has already been rejected.
-> But if I improve/modify the image file, I consider it to be a new submission

So far I have achieved an acceptance rate of 99.6% by submitting the same image several times - after 5 attempts I stopped, but until then almost every image went through.

In future, I will collect the rejections and increase the brightness by 5% once a month using batch processing and submit the whole lot again.

In the end, Shutter shows us through its own behavior that we have to find other ways to get our sales ^^

69
Adobe Stock / Re: What's your weekly ranking and how many images?
« on: November 29, 2024, 10:27 »
I am enjoying every moment for as long as the fun lasts.

I really wish I could get this income all year round.
And I am sure that you will succeed. You look very closely at your figures and your images, at niches that might open up.
That will definitely pay off in the long term - now you basically only need material every season and you have good sales for every season :)

At least that's what I keep thinking - diversifying the portfolio further will eventually result in more constant downloads - that's the hope :-)

70
Adobe Stock / Re: What's your weekly ranking and how many images?
« on: November 29, 2024, 07:46 »
This week is basically a total failure for me at AS...

It started well with 41DL on Monday, 25DL on Tuesday were still ok and from Wednesday it went downhill steeply with Wed 17 and Thu 12 DL - today on Friday I have exactly 1 DL at lunchtime - normally it's between 12 and 18 at this time of day.

Whatever AS has changed again (probably on Tuesday afternoon), for me this means that it's going down even further than it has been since March 24 - and that with constant uploads every day.

I just love this time of the year
pos 856, files 7050
I'm very pleased to read that :-) Congratulations!


71
Adobe Stock / Re: What's your weekly ranking and how many images?
« on: November 27, 2024, 06:55 »
Made it into the inner 1k. Happy day :)
I am really happy to see this!!
Congrats and Herzlichen Glckwunsch :-)

72
Shutterstock.com / Re: Video submission limits?
« on: November 23, 2024, 17:30 »
If I'm not completely mistaken, there is a general limit per week for the upload. I think it was 100 videos and 500 photos?
But I'm not sure about the numbers - it's been a while since I had such a message.

73
Adobe Stock / Re: Free Collection (Video)
« on: November 22, 2024, 06:54 »
Hmm. Yeah. I had the teaser of close to that amount... and then... nada. While I didn't (necessarily) expect ALL of them to be accepted, I thought maybe around 10% or so, so thought maybe I'd get a couple grand... instead... nada. Hmm...
Hmm... with about 1.500 nominated you got 0 selected?
That is strange indeed...

74
Adobe Stock / Re: Free Collection (Video)
« on: November 21, 2024, 18:12 »
They approved 36 of 1451 nominated in my case - just the average 2-3% of the nominated numbers.

That message telling you (in my case): "You can earn up to $11.450..." is just a teaser to make you click: Yes send my files to the free collection - and thats fine i guess.

75
123RF / Re: 504 Gateway Time-out on contributor
« on: November 20, 2024, 14:13 »
No, you are not the only one with the time-out messages.

@Stoker:
Just because it works for you, to conclude that Louis is the only one with the message is, to put it mildly, a bold statement ;-)

For weeks/months now, 123rf has been in a mess.
The site often doesn't work for contributors, the ftp upload hasn't worked properly for a while (the ftp certificate expired years ago) and the reviews haven't worked for months (4,200 images in the queue since February 24).
On average, support only replies to every 3rd email and then only with a standard reply from the archive.
It looks to me very much like the next agency will soon be closing its doors here...

Pages: 1 2 [3] 4 5 6 7 8 ... 21

Sponsors

Mega Bundle of 5,900+ Professional Lightroom Presets

Microstock Poll Results

Sponsors