pancakes

MicrostockGroup Sponsors


Author Topic: Adobe Stock generative AI reminders  (Read 22412 times)

0 Members and 2 Guests are viewing this topic.

« Reply #50 on: August 26, 2023, 08:25 »
+4
As I was looking through all those AI generated stock photos, I realized 70-80% of stock photos as we know can be generated by AI.  So, why bother to shoot stock photos anymore unless it's editorial news photos, I thought.  It's over for real camera shooters.  AI generators don't have to hire models, travel to locations and setup lightings.  Can't compete against those.
And further, why would anyone buy an AI generated stock image, when they could just as well have their own unique image AI generated?


« Reply #51 on: August 26, 2023, 08:44 »
+2
As I was looking through all those AI generated stock photos, I realized 70-80% of stock photos as we know can be generated by AI.  So, why bother to shoot stock photos anymore unless it's editorial news photos, I thought.  It's over for real camera shooters.  AI generators don't have to hire models, travel to locations and setup lightings.  Can't compete against those.
And further, why would anyone buy an AI generated stock image, when they could just as well have their own unique image AI generated?

It's easier and cheaper to buy on Adobe Stock or Shutterstock on cheap subs. 

« Reply #52 on: August 26, 2023, 09:06 »
+3
As someone who has been doing ai generation daily for monthsit is a lot faster to create content with a camera.

Especially if you work with real people, you get a huge amount of files in one afternoon that are instantly usable, need little postprocessing and you can do videos at the same time.

And you get personalized images. With ai you have the problem that if somebody else is using the same prompt, they will get a duplicate or extremely similar image to yours.

As for travel imagesthe images are not accurate. You can add a location, but is that cactus growing in the desert really from Mexico?

In my underwater tests I got all kinds of new fish species.

If you try to prompt local landmarks (I know we should not do this for Adobe) my cathedral suddenly has three spires instead of two, the bridges all look wrong

The ai is creative it mixes and matches just on visual similarity.

It is not a fact checker and absolutely not accurate.

So no, I see ai content as an interesting addition, but mostly I see it as a good photoshop tool.

For instance if I do studio christmas cards in the snow, i can then use ai to easily overlay snow and lighting filter effects, adjust colors or add a quick vintage look.

So I can expand on my photos.

With people you can add or remove beards, change the daytime and maybe even change ethicity and skin color in a realistic way.

Then you could take a group of white caucasians and transform the group into a more diverse mix, even different for different countries - more mideast people for Germany, more black or hispanic for the USA

It is a tool.

If you only prompt with ai without your own base photos, or without a mile long complex prompt, you will just end up with an endless duplicate midjourney port.
« Last Edit: August 26, 2023, 09:15 by cobalt »

« Reply #53 on: August 26, 2023, 09:08 »
+1
As I was looking through all those AI generated stock photos, I realized 70-80% of stock photos as we know can be generated by AI.  So, why bother to shoot stock photos anymore unless it's editorial news photos, I thought.  It's over for real camera shooters.  AI generators don't have to hire models, travel to locations and setup lightings.  Can't compete against those.
And further, why would anyone buy an AI generated stock image, when they could just as well have their own unique image AI generated?

I've looked at a huge number of AI generated images accepted at Adobe Stock, and I think the reason that people are buying, and probably will continue to buy, human-produced stock images is that a very large number of AI generated images aren't usable.

Impossible staircases in luxury interior shots, ladders you can't climb, kitchens with door handles at all angles, stools missing legs or sitting at bizarre angles, people with three thumbs, three legs or missing some body parts, hammocks suspended in thin air, table lamps growing out of books, doors you can't get to - it goes on and on and it isn't getting better with newer submissions.

Buyers can't use these except as novelty items or to create memes. My experience with images that sell suggests that there are lots of real world businesses that need real-world images for their marketing materials and web sites.

And then there are the people. By and large they look artificial - beyond any overdone retouching we've typically seen in stock shots. There may be a niche market for a small number of these, but I don't see this stuff going mainstream.

If you risk eyeball damage by looking over what Shutterstock has for AI generated images (what customers made with their Dall-E based tool) you'll understand why in the earnings call SS said that they saw lots of experimentation but few downloads. They expected that would improve when the quality increased, effectively acknowledging the quality problem they have.

Firefly is still in beta but widely available now via Adobe Express. Reviews of that earlier this month mentioned the poor quality of results (which my testing of the beta would agree with)

Fantasy content seems to be where AI does best - because there are no rules. It's where it intersects with the real world that it has trouble - and that's where a huge segment of stock licensing operates.

Edited to add: I just went to look at new genAI uploads and the first two images were of a kitchen. Just look at all the errors in this image (2nd one) - freshly approved...



The stool legs are missing parts of their supports; the cabinet handles are all over the place; stovetop knobs are mashed pixels; the stool on the far side of the island has mangled legs; there's a light cord on the left but no light; the fridge doors are missing handles - and that's just what I can see in the preview image. This is useless and should not have been approved.
« Last Edit: August 26, 2023, 09:34 by Jo Ann Snover »

« Reply #54 on: August 26, 2023, 09:17 »
0
I agree, for fantasy content, but also for anything art related - paintings, watercolor, charcoal, clipart, papercraft it is amazing what you can generate.

But as soon as you need the accuracy of the real world, it is very difficult.

I am now trying to do people and food, genres that need very good accuracy, and it is extremely difficult to get good results.


« Reply #55 on: August 26, 2023, 10:06 »
+2
Contrary to the advice here, I'm doing photorealistic stuff. I'm doing it exactly because it's hard, that means fewer producers can't make it, and it's not as simple as putting a few words in MJ and calling it a day.

Just a few days ago I had a genAI sale of a person for over $30.

I'm not afraid of the "competition" using midjourney - skin looks fake, people are in the uncanny valley, and we're going to laugh about pictures like that in a few months just like today we're laughing when we see "person isolated on white" shots.

Of course, producing high-end genAI takes time, but imho as long as Adobe cleans up their library, it's worth it.

« Reply #56 on: August 26, 2023, 11:48 »
0
You can do photrorealistic stuff, but you say yourself it is hard.

Compared to one afternoon with real people, what will get you more files, more choices, more scenes and video?

I see an advantage if you are trying to do engineers working on an oil plattform, medical doctor performing open heart surgeryall kinds of things where it is difficult to get access to or to get releases.

If you can generate that with ai, with very good detail accuracy, then yes, I think ai will have the advantage, because simply the time you need for organising and preparing that kind of shoot is very time consuming and good models/actors also costs quite a lot.

I am determined to learn my way around people and food, but it certainly is a much bigger challenge than easter eggs.

« Reply #57 on: August 26, 2023, 13:39 »
+3

I see an advantage if you are trying to do engineers working on an oil plattform, medical doctor performing open heart surgeryall kinds of things where it is difficult to get access to or to get releases.



Cobalt, that's exactly what I tried and didn't get anything usable if you're interested in photorealistic detail. This is somehow understandable, because in many areas where it is difficult to get access for a photographer,  there are not many reference images to train the AI as well.

For all the medical subjects I had in mind, the AI failed completely.

Even with classic simple doctor images with stethoscope around the neck, the stethoscope usually had ear pieces on both sides.
Our new Youtube trained generation of AI contributors should not care about that - especially because the pictures looked superficially pretty at first sight  - but stethoscopes are not designed for whispered mail  ;)
« Last Edit: August 26, 2023, 14:33 by RalfLiebhold »

« Reply #58 on: August 27, 2023, 13:29 »
+1
...
I see an advantage if you are trying to do engineers working on an oil plattform, medical doctor performing open heart surgeryall kinds of things where it is difficult to get access to or to get releases.
....

awhile ago i asked for a surgeon doing brain surgery, while results showed diversity, one had a woman working on an isolated brain that looked more like a turkey

and my request for Darwin writing in his garden ( for a blog post), it showed how industrious the man was - he could write on both pages w a pen in each hand

wds

« Reply #59 on: August 28, 2023, 08:18 »
+1
Another question is how long will it take for AI to get beyond the non-sense/inaccurate images? Could be that it will be "quick" or could be it will take a long time due to the way AI image generation works.

« Reply #60 on: August 28, 2023, 08:24 »
+5
You never know but to create technically or medically accurate images...or just accurate fish species underwater and correct corals for a specific location...I think that will take a very, very long time.

It would have to understand entire ecosystems or entire professions including current details to be able to do that.

From working with ai...I sense nothing "intelligent" about.

Just a random visual mix, that gradually improves as users tell it ... no...not like this.

But I do not have the impression of the slightest bit of actual understanding what is in the image.

Working with ai makes me less scared of it.

« Reply #61 on: August 28, 2023, 15:14 »
+3
New approvals in the genAI collection continue to include some utterly useless, broken, messed up images. One portfolio (which is huge) had some major clunkers in recent approvals so I took a look at some of the rest.

It makes the point so clearly - cleaning up the review process is critical and should be the number one priority. Popups in the upload process won't do the job.

Just a small taste of what I'm referring to:














« Reply #62 on: August 28, 2023, 16:02 »
+5
New approvals in the genAI collection continue to include some utterly useless, broken, messed up images. One portfolio (which is huge) had some major clunkers in recent approvals so I took a look at some of the rest.

It makes the point so clearly - cleaning up the review process is critical and should be the number one priority. Popups in the upload process won't do the job.

Just a small taste of what I'm referring to:






Jo Ann,

I think you see the issue a bit too narrowly and too critically.

If a train travels faster than its CO2 emissions, then the zeitgeist is well met and highly topical  ;)

Thank you, the picture was my ;D highlight and is hard to top.

« Last Edit: August 28, 2023, 16:07 by RalfLiebhold »

« Reply #63 on: August 28, 2023, 16:13 »
+1
New approvals in the genAI collection continue to include some utterly useless, broken, messed up images. One portfolio (which is huge) had some major clunkers in recent approvals so I took a look at some of the rest.

It makes the point so clearly - cleaning up the review process is critical and should be the number one priority. Popups in the upload process won't do the job.

Just a small taste of what I'm referring to:













No worries.  Buyers will fix them to make them make sense.  At least these are not copyright/trademark violations.  Reviewer AI clearly cant distinguish 6 fingers, 3 arms, 3 legs and other weird images.
« Last Edit: August 28, 2023, 16:29 by blvdone »

Just_to_inform_people2

« Reply #64 on: August 28, 2023, 16:24 »
0
New approvals in the genAI collection continue to include some utterly useless, broken, messed up images. One portfolio (which is huge) had some major clunkers in recent approvals so I took a look at some of the rest.

It makes the point so clearly - cleaning up the review process is critical and should be the number one priority. Popups in the upload process won't do the job.

Just a small taste of what I'm referring to:






Jo Ann,

I think you see the issue a bit too narrowly and too critically.

If a train travels faster than its CO2 emissions, then the zeitgeist is well met and highly topical  ;)

Thank you, the picture was my ;D highlight and is hard to top.


Not everybody knows the german language here Ralf. And then the joke is not quite understood.

And it's quite bitterful that someone is having a laugh about AI while so many suffer of it's consequences and see their hard work and livelyhood be destroyed.

« Reply #65 on: August 28, 2023, 18:11 »
+2

No worries.  Buyers will fix them to make them make sense.  At least these are not copyright/trademark violations.  Reviewer AI clearly cant distinguish 6 fingers, 3 arms, 3 legs and other weird images.

You mentioned trademark violations? Today's approvals have those too...


« Reply #66 on: August 28, 2023, 18:13 »
+1
@RalfLiebhold

Somehow I'd never heard of the ability to outrun one's CO2 emissions :)

« Reply #67 on: August 28, 2023, 19:40 »
+1

No worries.  Buyers will fix them to make them make sense.  At least these are not copyright/trademark violations.  Reviewer AI clearly cant distinguish 6 fingers, 3 arms, 3 legs and other weird images.

You mentioned trademark violations? Today's approvals have those too...



I blame Apple. 

« Reply #68 on: August 29, 2023, 00:50 »
+2
New approvals in the genAI collection continue to include some utterly useless, broken, messed up images. One portfolio (which is huge) had some major clunkers in recent approvals so I took a look at some of the rest.

It makes the point so clearly - cleaning up the review process is critical and should be the number one priority. Popups in the upload process won't do the job.

Just a small taste of what I'm referring to:






Jo Ann,

I think you see the issue a bit too narrowly and too critically.

If a train travels faster than its CO2 emissions, then the zeitgeist is well met and highly topical  ;)

Thank you, the picture was my ;D highlight and is hard to top.


Not everybody knows the german language here Ralf. And then the joke is not quite understood.


This is called germanism.
As far as I know, the term zeitgeist is used as a loan word in the English language in the same way as, for example, kindergarten or German Angst.
But I'm happy to be taught better.
« Last Edit: August 29, 2023, 01:03 by RalfLiebhold »

« Reply #69 on: August 29, 2023, 01:19 »
+1

No worries.  Buyers will fix them to make them make sense.  At least these are not copyright/trademark violations.  Reviewer AI clearly cant distinguish 6 fingers, 3 arms, 3 legs and other weird images.

You mentioned trademark violations? Today's approvals have those too...



The office swivel chair is also interesting. The lack of a backrest indicates an ergonomic innovation.

« Reply #70 on: August 29, 2023, 02:33 »
+1
Here is an interesting web page. You can type your name or image and see if your images are used as training. Some of mine are there. I hope someone will go to court to stop this madness.

https://haveibeentrained.com/

« Reply #71 on: August 29, 2023, 02:40 »
0
As I was looking through all those AI generated stock photos, I realized 70-80% of stock photos as we know can be generated by AI.  So, why bother to shoot stock photos anymore unless it's editorial news photos, I thought.  It's over for real camera shooters.  AI generators don't have to hire models, travel to locations and setup lightings.  Can't compete against those.
And further, why would anyone buy an AI generated stock image, when they could just as well have their own unique image AI generated?

It's easier and cheaper to buy on Adobe Stock or Shutterstock on cheap subs.

But doing it yourself you can get exactly what you want.

« Reply #72 on: August 29, 2023, 03:06 »
0
I'm considering using AI-generated images as a reference for colors and composition, and then redrawing them using Adobe Illustrator with additional elements. Will upload it as Vector illustration. Do I still need to show that it is AI generated? Thanks

Someone asked Matt a the question "if the AI generated image is used as a sketch and heavily processed after, must we always write made with generative AI?" and the answer was yes.
Though, it's not like Adobe really has a way to know that you used an AI or even care, seeing as how many obvious AI images that are not labaled as AI are their database.....


Mat, We need a clear answer.

We do not accept generative AI vector images. I strongly advise against what you are suggesting. Using the tool as inspiration is one thing, auto-trace or something similar I would avoid at all cost.

Thanks for the question,

Mat Hayward

Redrawing is not autotracing.

« Reply #73 on: August 29, 2023, 03:12 »
0
As I was looking through all those AI generated stock photos, I realized 70-80% of stock photos as we know can be generated by AI.  So, why bother to shoot stock photos anymore unless it's editorial news photos, I thought.  It's over for real camera shooters.  AI generators don't have to hire models, travel to locations and setup lightings.  Can't compete against those.
And further, why would anyone buy an AI generated stock image, when they could just as well have their own unique image AI generated?


But doing it yourself you can get exactly what you want.

That is also true for photos and videos and yet we have sales every day. Even very mundane photos anyone can do quickly with a mobile phone.

Customers don't have time. We are a resource for them to work faster.

It's easier and cheaper to buy on Adobe Stock or Shutterstock on cheap subs.

« Last Edit: August 29, 2023, 07:07 by cobalt »

« Reply #74 on: August 29, 2023, 08:44 »
+2


The Apple logos just keep on getting accepted

Fix the review process


 

Related Topics

  Subject / Started by Replies Last post
234 Replies
34498 Views
Last post May 27, 2023, 12:12
by cobalt
10 Replies
2676 Views
Last post April 28, 2023, 00:15
by wordplanet
52 Replies
7412 Views
Last post July 13, 2023, 06:15
by Justanotherphotographer
18 Replies
2732 Views
Last post July 24, 2023, 12:32
by MxR
23 Replies
3916 Views
Last post December 14, 2023, 22:05
by synthetick

Sponsors

Mega Bundle of 5,900+ Professional Lightroom Presets

Microstock Poll Results

Sponsors