MicrostockGroup Sponsors


Author Topic: Ai creates Artwork - how to upload?  (Read 1263 times)

0 Members and 1 Guest are viewing this topic.

« on: July 02, 2022, 08:50 »
0
Hello together,

I am currently facing the question of how I should upload images generated by an AI to AdobeStock?

Brief background:
I program neural networks myself without external libraries and without external image data. These generate stylized images (see examples below).
(This is not that very nice Dall e 2-AI, it is completely my own)

Are such images photo or illustration?
Do I need a model release and where should I get it from, since there is no model?
Or a property release? I would then have to have my computer sign it...

What do you think?

And before I get into trouble, I thought I'd throw this question out there.

I would be interested in your opinion and of course an official statement, how Adobe would like it :-)

Michael


« Reply #1 on: July 02, 2022, 10:18 »
+1
I'd think this would be analogous to a cross between an illustration and a 3D render. Definitely not a photo.

If you use images of your own as reference for things like an eye, then that would work like a photo reference for an illustration. Property releases for your own photos used in other works and, depending on how much of the person is used in the generated image, possibly a model release for the person in the source photo.

I don't think there is anything currently in use at agencies to indicate you have the rights to the software that generated an image when it's not one of the commercial 3D products. Perhaps a property release from you for your own software?

« Reply #2 on: July 02, 2022, 11:00 »
+2
@JoAnn:
Thanks for your answer.
The problem with the reference images is, that there isn't a single reference image creating the eye - it is the resulting formula of thousands of images training the AI to create an eye.

And yes, a property release stating that I am using a self-developed software would be the best I can do.

Maybe I should just try to upload one of those create images as illustration and see what happens? :-)

« Reply #3 on: July 03, 2022, 08:29 »
+2
If the images are created by referencing other images, you may have a copyright issue.  Be sure to do a reverse image search to check for visible similarities. 

« Reply #4 on: July 03, 2022, 09:53 »
0
If the images are created by referencing other images, you may have a copyright issue.  Be sure to do a reverse image search to check for visible similarities.
Thanks, a good idea to do a reverse search to see if there are similar images.

The AI creates images from my very own images database and the results are a mathematical mixture of parameters of that images, extracted and transported via a neural network to a new created image.

« Reply #5 on: July 03, 2022, 10:28 »
+3
Hi,
You produce an image that is made from other images so first all images must be yours.

Example: let's say you want to make an landscape of an ocean in a desert with a moon. if you use Photoshop for that i am sure you know that the picture of desert, moon and ocean must be yours first. 

You don't need a property release of AI Gan technology SW like you don't need a property release of Photoshop SW to produce digital images.
What you need is just to Describe property covered by release including intellectual property, details of any registration / other description.

Something like Gan technology/digital image and reference to your intellectual property number is enough.
If you are going to "make" new humans you must add all property releases (probably the "thousands of images" as you described, at least). So the human eye that you generate with GAN it is arguably. It is not generated with a 3d SW like c4d, blender, Maya or Zbrush but with AI-Gan since the stitches, blur and merge are an evidence in the image (typical artifacts from GAN merge pictures) not to mention the missing of eyelashes.

« Last Edit: July 03, 2022, 10:57 by Evaristo tenscadisto »

« Reply #6 on: July 03, 2022, 12:52 »
0
... So the human eye that you generate with GAN it is arguably. It is not generated with a 3d SW like c4d, blender, Maya or Zbrush but with AI-Gan since the stitches, blur and merge are an evidence in the image (typical artifacts from GAN merge pictures) not to mention the missing of eyelashes.
Thanks for your answer - it seems to be a complicated case, all this AI stuff will for sure be a problem in the near future.

What I would like to point out is, that the eye is not a stitched images (sorry, i hope i can find the right words - my english is not that good). The eye is generated by a formula trained by lots of eyes, but to generate this eye, not a single pixel of the original images is used.
It is just a formula with about 70.000 variables. The "stitches" are mathematical gradient boundary of the non linearity in the formula and not edges of images stitching.

« Reply #7 on: July 06, 2022, 01:00 »
+1
Justanimage, I understand what you are saying.

Please understand also that all digital images are made in a unique combination of RGB pixels which is transformed in to numbers. Each Digital Image is combine in a unique way that may depend in the format you use (tga,png, jpeg, Gif, etc)- Mathematical formulas in binary code. The image itself is protect by law digital (binary code not pixel) or analogue and you can't use it even for training without consent of the owners.

In 2004 i was in R&D project where we need to train an algorithm to improve the pap smear diagnosis for the doctor based on image match. So you can imagine the process... a lot of diplomacy talk, European commission, Project officers, Evaluators, lawyers, Signatures, paperwork and consent (women, hospital and government)

But let's do the other thinking to explain it better...an image is binary code. your formula (code) too. So imagine you release your formula/image online so people can purchase or see your work. Now imagine that i will take it and put in a machine learning to the point where i create a new one. I will be violating your own work, right? Even if it was new the binary code was once based on yours or not? you know it was... otherwise will be a blank code producing blank formulas.

Actually this thinking already exists and its called "machine dreaming" and it's made by steps through generative adversarial network (GAN-AI) to produce new formulas(binary code), images (png, jpg, mp4) or sounds (wav, mp3).

So you said   "The eye is generated by a formula trained by lots of eyes....."  therefore i must ask,  real people eyes or "made" by you? do you have consent? do you ask or pay?

"... but to generate this eye, not a single pixel of the original images is used."  Of course not. You don't need it anymore. you train your machine to produce the necessary binary code/formula with generative adversarial network.

I will give an good example of my point. there is the site called https://generated.photos/

Please go there.... it is a microstock site where you can buy Generated photos like yours created from scratch by AI systems. All images can be used for any purpose without worrying about copyrights, distribution rights, infringement claims, or royalties. But you must pay to use....of course

and at the end of site you will see that was trained on studio photos by Icons8 who was paid or give consent from owners, right? 






 
« Last Edit: July 06, 2022, 01:16 by Evaristo tenscadisto »

« Reply #8 on: July 06, 2022, 05:29 »
0
Hello Evaristo,

Thank you for the detailed answer. You obviously have a deeper understanding of neural networks than I do - I've only been working on programming them for about a year in my own environment (without Tensorflow or Keras).

Short answer to your questions: Yes, I train the networks exclusively with my own photos that don't require rights clearance or of which I have the rights clearance - it's just hard to add a few thousand reference photos as attachments when uploading :)

The website is a great example and I am familiar with the GAN networks - but my own NN environment is not yet that far advanced.

I'll wait and see how things develop with the "AI images" on the stock market - something should happen very soon, because I think the topic is becoming very acute.

For the creation of such images, I don't want to be dependent on another system later on, so that I don't have to ask for permission - hence the programming of my own environment.

Translated with DeepL.com

« Reply #9 on: July 06, 2022, 08:00 »
+2
The AI is a tool like any other tool.  You use it to create images.  Youve used your own work to train it.  I dont see why you need any releases uploaded at all.  And afaik, an agency only calls it a photo if it was created pixel by pixel and an illustration if created with vector software.  So, a photo.
« Last Edit: July 06, 2022, 14:26 by Sean Locke Photography »

k_t_g

  • Professional Dreamer
« Reply #10 on: July 06, 2022, 22:54 »
0
Neat tool but I only see this tool creating "good enough" avatar graphics and not something you would sell on MS. Besides wouldn't that be cheating/stealing? I mean you are using other people's work.

« Reply #11 on: July 07, 2022, 09:10 »
0
Neat tool but I only see this tool creating "good enough" avatar graphics and not something you would sell on MS. Besides wouldn't that be cheating/stealing? I mean you are using other people's work.

Did you not read the OPs posts?

« Reply #12 on: July 07, 2022, 13:48 »
0
Justanimage, I understand what you are saying.

...


I will give an good example of my point. there is the site called https://generated.photos/

Please go there.... it is a microstock site where you can buy Generated photos like yours created from scratch by AI systems. All images can be used for any purpose without worrying about copyrights, distribution rights, infringement claims, or royalties. But you must pay to use....of course

and at the end of site you will see that was trained on studio photos by Icons8 who was paid or give consent from owners, right? 


They are scarily natural looking ... but interesting that they claim the images are "high resolution" when the largest image size available is 1024 x 1024 pixels.

Perhaps the OP can put together a property release that says AI images are based on a GAN (describe it and say it's his software) trained by using thousands of 1,000s of model released images taken by him? Perhaps they'd accept a contact sheet with example images since you can't upload 1,000s?

It may require getting in touch with Adobe to ask them what they require. Maybe Mat can direct you to the right person to contact at Adobe. It's not an easy question to answer but this is obviously something that will come up more and more so the agencies will need to determine what is required by way of releases.

When AI can generate people and scenes that are really high definition we will be in trouble. But for scenes they will need to license 1,000s of images for the machine learning - and microstock gives them easy access to those photos - do they license them as artist reference? Lots of interesting questions.


« Reply #13 on: July 07, 2022, 14:26 »
0

"When AI can generate people and scenes that are really high definition we will be in trouble."

you already can! Just use Ai scale for that. You can purchase for 99$ Gigapixel from Topaz labs products you can try for free too. I've been using it since 2019/2020.  With this Ai sw you can Upscale and enhance any image by up to 600% while preserving image quality, texture, and detail.

But my major concern is actually using eyes for training even in academic purpose. You see there's not just AI art going on. And the eyes just like fingerprints, identical twins don't share the same iris swirls and patterns, so each of their irises is also unique.

I am sure you all are aware of AI tracking. There is a lot of types of Ai tracking... you can have already also in cameras such as R5, right? So "imagine" that instead of using the AI pyton code to use 70K eyes and produce a new unique eye you use it to track similar images over the net. This is not sci fi anymore. I think we all know there is a device that we use and look at all time and constantly doing selfies and tik tok movies, shared in social media, right?

It is not distopic to think that AI in the wrong hands can have bad results.

Anyway... there is always a workaround: use sunglasses! ehehehe 8) 



 

Related Topics

  Subject / Started by Replies Last post
29 Replies
13820 Views
Last post October 31, 2009, 17:43
by PeterChigmaroff
4 Replies
3630 Views
Last post August 27, 2012, 16:40
by leaf
189 Replies
27440 Views
Last post July 10, 2013, 21:29
by EmberMike
10 Replies
3007 Views
Last post May 22, 2014, 23:44
by Will
0 Replies
1752 Views
Last post February 07, 2020, 17:52
by umdash9

Sponsors

Mega Bundle of 5,900+ Professional Lightroom Presets

Microstock Poll Results

Sponsors

3100 Posing Cards Bundle