pancakes

MicrostockGroup Sponsors


Author Topic: Generative AI updates from Adobe VP Matt Smith  (Read 1420 times)

0 Members and 1 Guest are viewing this topic.



« Reply #1 on: April 23, 2024, 11:34 »
0
Thanks for the update Mat, looking forward to see how we can pave a way forward for both the creators and the technology spectrum.

« Reply #2 on: April 23, 2024, 12:09 »
+1
interesting read.thanks.

For my part,as I have already reiterated several times,I am more than sure that Adobe has made every effort to build Firefly in the most ethical way possible,respecting all parties involved.

i am not an AI expert,but I've generated about 20,000 images between Firefly and other AIs and from the results obtained,the difference seems clear as day to me,I have never found anything in the contents generated by Firefly that could even suggest that content was used incorrectly for AI training,while unfortunately,I cannot say the same thing about Midjourney,as too often there are pieces of logos,watermarks or signatures of authors in the generated content,so even a 6 year old child would understand that they used content inappropriately for AI training.

thank you for all the efforts made,like FAIR act,in trying to build a better future for us artists,contributors,in a very complex era of important changes.

keep it up Adobe! :)






ADH

« Reply #3 on: April 23, 2024, 12:39 »
+2
The days of stock images as a means of subsistence are numbered, we are much closer to the end than we imagined.

« Reply #4 on: April 23, 2024, 13:27 »
+4
The days of stock images as a means of subsistence are numbered, we are much closer to the end than we imagined.

even before AI appeared, many us held this was the case - witness the stories in the 'days of yore' thread.

 the 'end' is behind us

« Reply #5 on: April 23, 2024, 13:34 »
0
does this mean AS uses none of its editorial collection in training?

perhaps because AS has a stricter definition of 'editorial' th an some other agencies, it accepts some images as commercial that others would label editorial (eg, some crowd scenes, or images with identifiable people in the background.

« Reply #6 on: April 23, 2024, 14:09 »
+3
While firefly might be the most ethical - it is still trained off of images made with programs that were trained with unethically sourced data - scraping the entire web. So second generation unethical. Also the first round of training used the AS collection without asking the artists. After that there was an opt out - but the data had already been used. Legally they might have covered enough bases, but ethical - well, sort of.

If it doesn't produce images with logos that is a testament to how well they curated the images they used to train - not the ethics of it. Still an improvement, but not necessarily ethical.

Adobe is in the interesting position to be both trying to sell to the artists that create content and make them obsolete with generative AI  - or at least convince investors they can do both.

« Reply #7 on: April 23, 2024, 14:58 »
+2
the images Adobe used came from Adobe's collection not Midjourney,and have an owner who was paid,then whether they generated other images from Midjourney,to train Firefly,we don't know,but even in this case it is not a stolen image,but the product of a stolen image,therefore a completely different image from the stolen one,stolen not by Adobe but by Midjourney.

There is already a huge difference in this.

Midjourney literally stole content for its own training,without even bothering to remove logos,signatures,watermarks,not paying anything to anyone,and asking substantial sums for subscriptions.

Firefly paid for all the content used,they didn't steal anything from anyone,and they didn't upload content with logos,signatures and watermarks,also leaving free access to the platform for months.

There is a completely different approach from Adobe in every thing and circumstance,it seems clear to me.

« Reply #8 on: April 23, 2024, 15:05 »
+2
One thing I wanted to highlight,I think there should be more clarity on when to label content as AI.

Maybe I missed something,since at the moment I'm more focused on real content,but for example,with the real subject and the AI ​​generated background,should the content still be labeled as AI?

If I correct only part of the image using generative fill,leaving the subject and much of the image real,do I still have to label the content as AI?

here all this is not very clear to me,and I think many are asking themselves the same question,and up to now I have only created completely AI or completely real content,to be safe and not make mistakes.

Even with just one AI-generated pixel,do I still have to label the content as AI? :D

« Reply #9 on: April 23, 2024, 15:27 »
+1
...

Even with just one AI-generated pixel,do I still have to label the content as AI? :D

don't underestimate single pixels --after all, the sophons are omniscient proton-sized nanobots

« Reply #10 on: April 23, 2024, 16:51 »
0
...

Even with just one AI-generated pixel,do I still have to label the content as AI? :D

don't underestimate single pixels --after all, the sophons are omniscient proton-sized nanobots

the 3-body problem!:D ok you've definitely already seen fallout! :) I just recently saw that the 5th season of StarTrek Discovery is out,later I'll start the free trial via Amazon Prime,since i have already used the free trial in Paramount! :D

anyway yes,I wondered if something had changed since Mat's last livestream,however it would be interesting if we could have a checkbox for content where we can declare that the subject is real,because it's a shame to have to label content as AI if only an abstract background is added for example,who knows maybe in the future! :)

« Reply #11 on: April 23, 2024, 18:27 »
0
I love this sentence.

"We believe that generative AI technology should support creators and allow them to monetize their talents. "

« Reply #12 on: April 23, 2024, 18:33 »
+4
...

Even with just one AI-generated pixel,do I still have to label the content as AI? :D

don't underestimate single pixels --after all, the sophons are omniscient proton-sized nanobots

the 3-body problem!:D ok you've definitely already seen fallout! :) I just recently saw that the 5th season of StarTrek Discovery is out,later I'll start the free trial via Amazon Prime,since i have already used the free trial in Paramount! :D

anyway yes,I wondered if something had changed since Mat's last livestream,however it would be interesting if we could have a checkbox for content where we can declare that the subject is real,because it's a shame to have to label content as AI if only an abstract background is added for example,who knows maybe in the future! :)

Here is the official policy per our learn and support page on generative AI about when to tag as AI and when it is not necessary.

Does the use of Generative Fill and Generative Expand in Photoshop and Generative Recolor in Illustrator require labelling an image as generative AI?

Label your image as generative AI when the use of generative AI tools in Photoshop or Illustrator changed, augmented, or added a new primary subject of an image.

Examples when to label your image as generative AI:

Generating new additions, such as a new person, animal, or object
Adding new subjects might compromise the accuracy of an image
Making significant adjustments or changes to a human subject in an image
Recoloring the primary subject of an image

Examples when you wouldnt be obligated to label an image as generative AI:

Extending background for any reason
Removing IP or other forms of retouching
Removing distracting objects or people
Recoloring the background of an image

« Reply #13 on: April 23, 2024, 19:36 »
+2
...

Even with just one AI-generated pixel,do I still have to label the content as AI? :D

don't underestimate single pixels --after all, the sophons are omniscient proton-sized nanobots

the 3-body problem!:D ok you've definitely already seen fallout! :) I just recently saw that the 5th season of StarTrek Discovery is out,later I'll start the free trial via Amazon Prime,since i have already used the free trial in Paramount! :D

anyway yes,I wondered if something had changed since Mat's last livestream,however it would be interesting if we could have a checkbox for content where we can declare that the subject is real,because it's a shame to have to label content as AI if only an abstract background is added for example,who knows maybe in the future! :)

Here is the official policy per our learn and support page on generative AI about when to tag as AI and when it is not necessary.

Does the use of Generative Fill and Generative Expand in Photoshop and Generative Recolor in Illustrator require labelling an image as generative AI?

Label your image as generative AI when the use of generative AI tools in Photoshop or Illustrator changed, augmented, or added a new primary subject of an image.

Examples when to label your image as generative AI:

Generating new additions, such as a new person, animal, or object
Adding new subjects might compromise the accuracy of an image
Making significant adjustments or changes to a human subject in an image
Recoloring the primary subject of an image

Examples when you wouldnt be obligated to label an image as generative AI:

Extending background for any reason
Removing IP or other forms of retouching
Removing distracting objects or people
Recoloring the background of an image

so, basically, common sense!

« Reply #14 on: April 24, 2024, 04:52 »
+1
ok Mat,so in fact something has changed in the meantime,from what I remember even the removal of a shadow,with the generative fill required labeling the content as AI.

so more freedom,well,adding a background generated with AI for example does not force you to label a content as AI,just like eliminate distracting elements with generative fill.

very interesting,I didn't know that,this shows how important it can be to be constantly updated on everything,things change quickly.

Is this also applicable to illustrative editorials?

thanks Mat! :)
« Last Edit: April 24, 2024, 04:57 by Injustice for all »

« Reply #15 on: April 24, 2024, 23:16 »
0
However Mat,if I can express my opinion,the explanation of when to label an image as AI and when not is too approximate,it needs to be improved,it is an important thing that can lead contributors to make mistakes and risk blocking or deletion of content with consequent waste of time.

For example,recoloring the background of an image:

so change the color but not replace with another background?

extending the background is fine?but any background?so if I have half AI and half real sea landscape as background is it fine?

I don't know if you get my point,but I don't think common sense is detailed enough,my common sense may be different from someone else's common sense,I think we need more details on this important differentiation.

I think an update on that page is needed.

Livestream maybe? :D


« Reply #16 on: April 25, 2024, 04:46 »
+1
and I also add,(tripling myself sorry) that look like there is a new category,which we can perhaps call "hybrid"

Hybrid content is half-AI,half-real content,apparently for a few months it has also been possible to create this type of content,therefore in my opinion these contents should be identified as such in some way.

subject or no subject,I believe that customers,and all parties involved,need to know that AI-generated elements are present in the content.

if we start mixing the cards too much we risk distorting reality a little too much,and I believe that it is our responsibility first of all to ensure that this does not happen.

this is my opinion,it's not my call,so the decision certainly isn't up to me,I'm just a simple contributor,but it seemed right to say it.

in the meantime I have already sent yesterday my first hybrid content,I want to see if it is accepted. :)


« Reply #17 on: April 25, 2024, 05:06 »
+2
Matt:  I've had illustrations rejected by Adobe because:

- Non compliant use of another artists name.
- Undeclared Generative AI Content

This is not true - I don't use and have never used AI.

Do I risk getting banned from Adobe from incorrect AI rejections?

Thanks
Cat

« Reply #18 on: April 25, 2024, 05:12 »
0
Matt:  I've had illustrations rejected by Adobe because:

- Non compliant use of another artists name.
- Undeclared Generative AI Content

This is not true - I don't use and have never used AI.

Do I risk getting banned from Adobe from incorrect AI rejections?

Thanks
Cat
Proof of their incompetence in handling situations. They are barely capable of giving a marketing speech, everything is fine, we are managing!
And the contempt for honest contributors from the start is growing, as they depend less and less on them. The contributor community is divided, great for them, they can rule.
« Last Edit: April 25, 2024, 05:32 by DiscreetDuck »

« Reply #19 on: April 25, 2024, 05:17 »
0
The days of stock images as a means of subsistence are numbered, we are much closer to the end than we imagined.
Sure

« Reply #20 on: April 25, 2024, 05:46 »
0

Quote
Proof of their incompetence in handling situations. They are barely capable of giving a marketing speech, everything is fine, we are managing!
And the contempt for honest contributors from the start is growing, as they depend less and less on them. The contributor community is divided, great for them, they can rule.

I still don't know how this will play out.  If people are not getting any compensation for their work - why would anyone post anything to the internet other than for fun.  If people stop posting anything high value such as artwork, photographs, video, written content what will AI work from - it's own outputs?  Do tech bros think people are going to spend their own dime and time for them to profit from? 

To be fair to Shutterstock - they rejected some of my images under "AI" but I raised a support call and it was sorted pretty quickly.


« Reply #21 on: April 25, 2024, 05:49 »
+2
Matt:  I've had illustrations rejected by Adobe because:

- Non compliant use of another artists name.
- Undeclared Generative AI Content

This is not true - I don't use and have never used AI.

Do I risk getting banned from Adobe from incorrect AI rejections?

This happens to me occasionally when I submit fractals and 3D renders and the reviewers think they are AI. Last time it happened I just re-uploaded the image and it was accepted. On other occasions I have written to Support with the file numbers and the files have been accepted. This is not the sort of thing that people get banned for.

« Reply #22 on: April 25, 2024, 06:28 »
+1

Quote
This happens to me occasionally when I submit fractals and 3D renders and the reviewers think they are AI. Last time it happened I just re-uploaded the image and it was accepted. On other occasions I have written to Support with the file numbers and the files have been accepted. This is not the sort of thing that people get banned for.

Thanks - reassuring

« Reply #23 on: April 25, 2024, 07:27 »
0
Scaling moderation capabilities by:
  • Doubling the size of our expert human moderation team who specialize in content review
  • Creating and training the team on new generative AI specific policies and best practices
  • Introducing specific generative AI moderation systems to efficiently categorize and review content for any potential policy violations
  • Refining our audit systems to monitor the existing collection more efficiently for content that may violate our new generative AI policies
  • Reviewing and removing thousands of existing assets from Stock and continuing to actively monitor the full Adobe Stock collection
So, the distance between a blog post and a dedicated forum is... one week.  ;)

'You' see here, they only talk about content, the contributor already no longer exists when it comes to moderation.
The future is autonomous Adobe, without (or with a minimum) of contributing suppliers, with less(no) royalties to pay, but an in-house team to manage the content and systems. And all customers will be so happy to feel like artists...
All this will be in their financial interest, it will inevitably happen!
AI content don't belong to contributors. Soon, AI content will no longer come from contributors.

The brainless prompters sheeps will realize too late, once they are ejected by autonomous systems and will come here to complain en masse. See you very soon!
They will still be able to apply for a job to join the "expert human moderation generation team" at Adobe... but it will be a small enclosure for a small herd.
« Last Edit: April 25, 2024, 07:58 by DiscreetDuck »

« Reply #24 on: April 25, 2024, 08:26 »
+2
@DiscreetDuck

please,here we are trying to understand how we should work,I have no time for these conspiracy theories and dystopian worlds.

instead let's try to understand for example how to behave or when to label AI content and when not,and what we can do and what not.

Let's try to ask questions that can be useful in our work rather,I'm trying to understand how I should work,thanks!


in the meantime my first hybrid content has been accepted!  :)




 

Related Topics

  Subject / Started by Replies Last post
10 Replies
2753 Views
Last post April 28, 2023, 00:15
by wordplanet
18 Replies
2818 Views
Last post July 24, 2023, 12:32
by MxR
185 Replies
23155 Views
Last post October 17, 2023, 02:37
by Deyan Georgiev Photography
5 Replies
1146 Views
Last post January 20, 2024, 08:35
by Deyan Georgiev Photography
2 Replies
424 Views
Last post April 16, 2024, 03:46
by Injustice for all

Sponsors

Mega Bundle of 5,900+ Professional Lightroom Presets

Microstock Poll Results

Sponsors