pancakes

MicrostockGroup Sponsors


Author Topic: Dall e 2 will make us all redundant?  (Read 4156 times)

0 Members and 1 Guest are viewing this topic.

« on: June 28, 2022, 04:12 »
0
Hi all. I've been blown away by how the A.I software Dall e 2 can make any image possible and when it goes live I just wonder what impacts it will have on the stock photo business. The future looks scary and it's fast approaching. I've always been one to look at the future and evolve my talents to fit. It's going to be kinda hard with Dall e around.
A friend of mine has early access to it and the results are just astonishing. Especially if it's in its infancy. Where this all goes is mind-blowing. Thoughts???

https://openai.com/dall-e-2/?labs
https://www.youtube.com/watch?v=yCBEumeXY4A
https://www.youtube.com/watch?v=0fDJXmqdN-A


« Reply #1 on: June 28, 2022, 04:15 »
0
It's scary but I'm kinda in awe of it. Although it will obviously have negative impacts on business I also think it can help us when we are coming up with ideas? It could give us a new twist on something we want to create?

« Reply #2 on: June 28, 2022, 05:02 »
+3
This came up before.

I believe it just recombines images that must be first inputted to the database. The photographer of ever component should be getting paid whenever any part of the image is used, same as if a human was Photoshopping together an image.

Also the results are mostly garbage. The YouTube videos cherry-pick stuff to hype it for clicks (or maybe as an ad for the creators).

If you search hard enough you will find actual scientific papers evaluating at it with actual results, they arent nearly as impressive.

« Reply #3 on: June 28, 2022, 05:43 »
+2
I beg to differ. A friend of mine has been allowed on the testing list and the results are mind blowing. I thought that maybe it just scrambles up artwork from images but it doesn't. It learns from an algorithm and creates an artwork 100% original. The lighting it creates and how it cosmetically structures the image is truly breathtaking. I'll see if I can upload some of his images.

His prompt was he wrote 'Wolf Mad scientist mixing sparkling chemicals in a Laboratory, digital art'
These were the result. NB. these images did not exist before and is not a collage of many images. this is A.I learning from text.
Also how the A.I understands light is crazy good. You can also do photorealistic images. Remember this is in its infancy too.

« Reply #4 on: June 28, 2022, 05:57 »
0

« Reply #5 on: June 28, 2022, 06:02 »
+1
It most certainly does use pre-existing images and edits/ distorts them to create the new images, just like a person in Photoshop. One giveaway: "OpenAIs policy prohibits publishing photographic images produced by DALL-E that show people who might be recognizable" i.e. if the photos composed have recognisable models.
« Last Edit: June 28, 2022, 06:16 by Justanotherphotographer »

« Reply #6 on: June 28, 2022, 06:02 »
+2
I beg to differ. A friend of mine has been allowed on the testing list and the results are mind blowing.

The analysis of DALL-E 2 is anything but mind-blowing though:
https://arxiv.org/ftp/arxiv/papers/2204/2204.13807.pdf

"for 5 out of the 14 prompts, at least one of the ten images fully satisfied our requests. On the other hand, on no prompt did all of the ten images satisfy our requests"

Only 5 out of 14 test images providing any satisfying result isn't blowing my mind. Also, if you take a look at the examples in this test: Even if the AI got it right and understood what was supposed to be displayed in the the images, the how is still problematic.
Look at this example of a pumpkin and a tomato. That the AI couldn't understand what was asked of it in most cases is one thing, but does any of these images look like high quality food photgraphy to you? Do you think any of these would have any real sale potential on microstock agencies and this is the kind of quality customers are looking for? Can you imagine images like this being used in an ad for a restaurant?



I think that AI will replace us at some point. Not just us, but a lot of jobs will be replaced by AIs and machines at some point. But I don't think we are quite there yet and won't be for a while.
Just look at all the graphic programs using AI technology, like sky replacement: It has been out there for years, and it's still highly flawed and needs manuall fix up in most cases. If not even something more simple, that has been in development and published for years, is functioning correctly, I don't think we will see high quality AI generated images without flaws for years. Even with Dall e 2 - Has anyone actually seen a zoomed in high resolution full size test image? How many weird flaws will you find then?
« Last Edit: June 28, 2022, 07:01 by Firn »

« Reply #7 on: June 28, 2022, 06:08 »
+5
Ah c**p. I was just doing some more reading and it looks like it does an internet image search pairing words with images to find the right ones to use. So this is very much a rights grab by the tech company. So if they get away with it, yes we will all be replaced i.e. have our work stolen off a glorified Google search and never compensated.

« Reply #8 on: June 28, 2022, 06:10 »
+2
I feel maybe we're in a bit of denial or maybe I see the future a little more clearly. Yes, it's not 100% foolproof yet and some results are totally off. But most stock images are simple and if someone can type in exactly what they want and get good results then I don't know why they would spend hrs looking for the right images when they can quickly make an original one using A.I in seconds?

I'd say in the next 5 to 10 years the stock industry will basically be a future version of Dall e. It's a no-brainer. We've just got to think out the box and how we can use this new tech to our advantage.

« Reply #9 on: June 28, 2022, 06:12 »
0
Ah c**p. I was just doing some more reading and it looks like it does an internet image search pairing words with images to find the right ones to use. So this is very much a rights grab by the tech company. So if they get away with it, yes we will all be replaced i.e. have our work stolen off a glorified Google search and never compensated.

From my understanding, there is a big lib of images with the words relating to the image. It then looks at the image and learns from the words and then creates a new image from the words. 100% original. I think I read that Google or Elon is the money behind all this.
« Last Edit: June 28, 2022, 06:16 by Cider Apple »

« Reply #10 on: June 28, 2022, 06:15 »
0
It most certainly does use pre-existing images and edits/ distorts them to create the new images, just like a person in Photoshop. One giveaway: "OpenAIs policy prohibits publishing photographic images produced by DALL-E" that show people who might be recognizable" i.e. if the photos composed have recognisable models.

I think that's because you can put famous people in poses or places they never actually were can can be extremely dangerous for obvious reasons. Apparently, you cannot create images of harm, abuse, famous people, naked, etc etc.

« Reply #11 on: June 28, 2022, 08:31 »
+2
This is technology.
You should always have an idea of what you want or you need, before to use any technolgy to create an image.

The impact will be the same as the digital camera for everyone. Big impact. But it will never replace the idea of what to shoot.

Also, I think very few clients have exact and clear idea of the images they need. They always need someone that can understand and translate the idea for them.

« Reply #12 on: June 28, 2022, 10:14 »
+5
It is very impressive, but it is using pre-existing images at some point in the creation process.

Photographers and illustrators need to be paid for the use of those images, unless they've been donated.

Inevitably, it'll be like Pinterest and other tech start-ups - steal and never say sorry.

Uncle Pete

  • Great Place by a Great Lake - My Home Port
« Reply #13 on: June 28, 2022, 12:30 »
+2
I feel maybe we're in a bit of denial or maybe I see the future a little more clearly. Yes, it's not 100% foolproof yet and some results are totally off. But most stock images are simple and if someone can type in exactly what they want and get good results then I don't know why they would spend hrs looking for the right images when they can quickly make an original one using A.I in seconds?

I'd say in the next 5 to 10 years the stock industry will basically be a future version of Dall e. It's a no-brainer. We've just got to think out the box and how we can use this new tech to our advantage.

And online, create your own image for a fee, it produces 20 results, the buyer picks the one they want. If this was available for artists, I'd want to use it.

As for perfect results, who cares? You get enough and define things properly and the AI should be able to understand. Give it confusing or vague orders and sure it will produce, confusing or wrong results. GIGO Like bad searches that produce bad results, it's the input and vocabulary that communicates what the AI will create.

Tomato and Pumpkin? What the heck is that? I mean you tell me, what it's supposed to mean? A combination, a hybrid? Side my side? Half and half? Tricking AI and then saying it's flawed by giving the input poor definitions, isn't really proving anything. And sure, AI is only as smart as what it's programmed to do and what it has learned so far. Software like this can continue to be updated and keep learning more.

It is very impressive, but it is using pre-existing images at some point in the creation process.

Photographers and illustrators need to be paid for the use of those images, unless they've been donated.

Inevitably, it'll be like Pinterest and other tech start-ups - steal and never say sorry.

For example, what if it uses Public Domain images? What if they do pay for use or buy the rights?

Yeah, you know that digital cameras will never be good enough to replace film. (ask Kodak) And Microstock will never replace real professionals and the traditional image agencies. So it's obvious that some AI like this won't ever replace people at home snapping images?  ::)

I feel maybe we're in a bit of denial

Maybe? Yet I don't think it will replace photos and artists entirely, but it could replace a large segment of easy on demand illustrative type images.


« Reply #14 on: June 28, 2022, 12:47 »
+2

Give it confusing or vague orders and sure it will produce, confusing or wrong results. GIGO Like bad searches that produce bad results, it's the input and vocabulary that communicates what the AI will create.

Tomato and Pumpkin? What the heck is that? I mean you tell me, what it's supposed to mean? A combination, a hybrid? Side my side? Half and half? Tricking AI and then saying it's flawed by giving the input poor definitions, isn't really proving anything.

Maybe you want to take a look at the analysis before commenting on it, then you'd see that the instructions were extremely specific.

Uncle Pete

  • Great Place by a Great Lake - My Home Port
« Reply #15 on: June 28, 2022, 13:00 »
+1

Give it confusing or vague orders and sure it will produce, confusing or wrong results. GIGO Like bad searches that produce bad results, it's the input and vocabulary that communicates what the AI will create.

Tomato and Pumpkin? What the heck is that? I mean you tell me, what it's supposed to mean? A combination, a hybrid? Side my side? Half and half? Tricking AI and then saying it's flawed by giving the input poor definitions, isn't really proving anything.

Maybe you want to take a look at the analysis before commenting on it, then you'd see that the instructions were extremely specific.

I did last time we had this same discussion? Some of the results are perfect and very useful. Negatives don't work well. Having a word, will lead the Dall-E to include that word. GIGO, better definitions, make better image results. No it's not perfect.

"The visual quality of the images is stunning. We were particularly impressed with the ability to
capture the top-down perspective we requested in Example 6. A commercial artist might have
trouble getting DALL-E2 to deliver the exact results that they or their client require; an amateur
looking for clipboard-like art with less strict expectations may well get something striking and
close enough to what is needed with very little effort.

DALL-E 2 is unquestionably extremely impressive in terms of image generation. The system
succeeds in applying many diverse artistic styles to the specified subject with extraordinary
fidelity and aptness, and capture their spirit: cartoons are light-hearted, impressionist paintings
are peaceful and evocative, photographs of everyday scenes are naturalistic, noir photographs are
subtly disturbing. Images in realistic styles are almost always physically plausible (we note
exceptions in example 9 and 12 below); images in non-realistic styles conform to the particular
norms of the style. Many of the images that have been published demonstrate DALL-E 2's
remarkable ability to create striking surrealist images, such as the half-human, half-robotic face

Some aspects of the system's language abilities seem to be quite reliable. If a caption specifies
only two or three objects, the system almost always shows all of them. If a caption specifies a
feature of an object, then the image will generally show that feature somewhere, though, as we
will discuss below, not necessarily on the correct object. The examples that have been published
elsewhere demonstrate that DALL-E 2 can reliably follow stylistic instructions (we did this in
only one of our experiments). In examples 7 and 14 below, the system reliably follows viewpoint
specifications, even though 7 requires a non-canonical view of the scene."


If you are only going to look for negatives, you will only see the negatives? Seems the review has lots of praise for the software as well.

"Caption: A tomato has been put on top of a pumpkin on a kitchen stool. There is a fork sticking
into the pumpkin. The scene is viewed from above."

Yes it pretty well failed. Their examples are example to trick the software, intentionally difficult.

« Reply #16 on: June 28, 2022, 13:09 »
0


I did last time we had this same discussion?
I can't recall having this discussion with anyone before?


Uncle Pete

  • Great Place by a Great Lake - My Home Port
« Reply #18 on: June 28, 2022, 15:29 »
0


I did last time we had this same discussion?
I can't recall having this discussion with anyone before?

Deja Vu all over again?  :)

https://www.microstockgroup.com/general-stock-discussion/dall-e2-ai-just-killed-stock-photography-and-illustration/msg576187/#msg576187

I would not worry about this too much, at least not for the next 10-20 or so years till the technology is really advanced enough to be really flawless enough to be useful.

I've already seen an AI that generates random realistic faces ( https://generated.photos/faces ) and I have seen a video about an AI that comppletely generates you any kind of realistic looking landscape ( https://www.youtube.com/watch?v=3OgthWZ982Y ) and yet it hasn't killed stock photography.

It will all look cool and that in a presentation video, but I don't think it would pass the reality check yet. Take sky replacements in programs like Photoshop or Luminar for example: I don't know about you, but whenever I try them, there are still some flaws that I would need to fix - Luminar for example always replaces the reflections in windows with the sky and has some strange clipping around trees sometimes - and that's just the sky. Imagine how many weird glitches you would have in a completely computer generated image that you would need to fix with some basic design skills. That's more something for designers who knwo their way around a graphic program, but not for the end customer (unless he has really low standartds). 
I think that, yes, one day we will get there and these computer generated "photos" will become a real thing and a threat to stock photographers. But I don't think we are there yet, not for a while.

But still I don't think this is the end to photography or us, and for simple illustrations, as long as it's not too specific or tricky, maybe the AI like DALL-E2 can be useful. I'd want to own it. Just think of all the things I could create. And apparently you can upload and edit your own photos as well.

What Dall-e can and cannot do, some more detailed flaws. But the can part is still pretty stunning to me.

https://www.lesswrong.com/posts/uKp6tBFStnsvrot5t/what-dall-e-2-can-and-cannot-do


« Reply #19 on: June 28, 2022, 16:21 »
0
Hi all. I've been blown away by how the A.I software Dall e 2 can make any image possible and when it goes live I just wonder what impacts it will have on the stock photo business. The future looks scary and it's fast approaching. I've always been one to look at the future and evolve my talents to fit. It's going to be kinda hard with Dall e around.
A friend of mine has early access to it and the results are just astonishing. Especially if it's in its infancy. Where this all goes is mind-blowing. Thoughts???

https://openai.com/dall-e-2/?labs
https://www.youtube.com/watch?v=yCBEumeXY4A
https://www.youtube.com/watch?v=0fDJXmqdN-A

This is really nothing new to me in terms of technology. Go search in github or deep ai. you have a lot of AI trials that you can test. In fact i already sold some stuff made with AI with my database. Also videos and animations.

Scary is what we can do with Metahuman plugin for Unreal engine. Deep fakes are very accurate with this technology.

« Reply #20 on: June 28, 2022, 18:57 »
0
I think... you will still have a demand for human work - but it will be different. In some ways it will be going back to the way it was - in that people will want to have an "original" human piece of art... or "Authentic" photos by so & so, etc.

Just like "brad pitt" has made a name for himself - and there are probably 1000's of other actors that are much better than "brad pitt" - people would still want to get their picture with him, etc. So even if the AI (which are actually ALGORITHMS, not "intelligence") got to the point where you could have it script an entire photo realistic looking movie ("technically" possible nowadays with things like blender3D, although a lot of work) - people can still sell their work.

I.e., look at poker. Technically you could have computer algorithms "outplay" pretty much any human - yet - humans still want to play against other humans. So the poker sites work hard on ensuring it is a "human only" thing.

Tech was developed to be used "by" humans. While of course there is a pyscotic element in society (i.e., the pyschos organzing the fakedemic the last 2 years designed to killoff as many ppl as possible to "replace" them with robots) - normal humans still want to interact with normal humans.

So I believe there will still be a demand - just like even though yes - many ppl use digital cameras - there are people who still like using film. It will just be 'different'. And you'll just have to reposition yourself.

« Reply #21 on: June 28, 2022, 19:55 »
0
It is worth watching, but I doubt it will replace us in the next 10 years.

But it is fun to use, I play with the nightcafe creator.
« Last Edit: June 28, 2022, 20:49 by cobalt »

« Reply #22 on: June 28, 2022, 22:06 »
+1
It is worth watching, but I doubt it will replace us in the next 10 years.

But it is fun to use, I play with the nightcafe creator.

I think part of the idea is not actually replace us but increase the volume of our work. 10 years ago i barely could do 10 pictures a day with all the steps of cleaning, some rotoscopy, colorize etc... Today with AI in photoshop or LuminarAI and some scripts i can easily do 40 in a day. But if we talk about cleaning for just green screen my personal record is 327 pictures in a day.

In ten years perhaps volume will be so easy to create digitally than a lot of individuals can actually have a personal microstock site store with millions instead of thousands. But for news or documentary stuff i think this will continue shy with no big changes

 

« Reply #23 on: June 29, 2022, 02:31 »
+5
I think there will still be a place for a human design or illustrator - we need someone to pick what is good or not, or what fits with a project that is being put together, but the original artist is certainly going to take a hit.  When doing a design project recently I was purchasing stock photos and had I had the opportunity to use Dall e2 to create the content I would have.  I thought that exact thought and tried to get access but no such luck.

« Reply #24 on: June 29, 2022, 07:24 »
0


I did last time we had this same discussion?
I can't recall having this discussion with anyone before?

Deja Vu all over again?  :)

https://www.microstockgroup.com/general-stock-discussion/dall-e2-ai-just-killed-stock-photography-and-illustration/msg576187/#msg576187

I would not worry about this too much, at least not for the next 10-20 or so years till the technology is really advanced enough to be really flawless enough to be useful.

I've already seen an AI that generates random realistic faces ( https://generated.photos/faces ) and I have seen a video about an AI that comppletely generates you any kind of realistic looking landscape ( https://www.youtube.com/watch?v=3OgthWZ982Y ) and yet it hasn't killed stock photography.

It will all look cool and that in a presentation video, but I don't think it would pass the reality check yet. Take sky replacements in programs like Photoshop or Luminar for example: I don't know about you, but whenever I try them, there are still some flaws that I would need to fix - Luminar for example always replaces the reflections in windows with the sky and has some strange clipping around trees sometimes - and that's just the sky. Imagine how many weird glitches you would have in a completely computer generated image that you would need to fix with some basic design skills. That's more something for designers who knwo their way around a graphic program, but not for the end customer (unless he has really low standartds). 
I think that, yes, one day we will get there and these computer generated "photos" will become a real thing and a threat to stock photographers. But I don't think we are there yet, not for a while.

But still I don't think this is the end to photography or us, and for simple illustrations, as long as it's not too specific or tricky, maybe the AI like DALL-E2 can be useful. I'd want to own it. Just think of all the things I could create. And apparently you can upload and edit your own photos as well.

What Dall-e can and cannot do, some more detailed flaws. But the can part is still pretty stunning to me.

https://www.lesswrong.com/posts/uKp6tBFStnsvrot5t/what-dall-e-2-can-and-cannot-do

I searched for a Dall e2 thread before posting this but the search didn't show anything. Weird, I'd have just voiced my opinions on the original. Maybe I needed to add the -?


 

Related Topics

  Subject / Started by Replies Last post
18 Replies
4938 Views
Last post November 11, 2010, 23:27
by luissantos84
11 Replies
4385 Views
Last post August 08, 2013, 02:12
by anjaliaroha
3 Replies
2251 Views
Last post May 28, 2014, 16:50
by gnirtS
19 Replies
2933 Views
Last post May 18, 2022, 05:21
by thx9000
13 Replies
983 Views
Last post August 16, 2022, 15:27
by derby

Sponsors

Mega Bundle of 5,900+ Professional Lightroom Presets

Microstock Poll Results

Sponsors

3100 Posing Cards Bundle