Banning AI art on PP

Unfocussed Mike said

Theimagebear said

If you create images with no model or there is one but you choose to keep their identity private (a growing thing) then thats not what here is about.

You mean if I have a model who is not on purpleport and won't tell you who they are?

I have plenty of work on here that is demonstrative of what I can do as a photographer, that features three people who enjoy working with me but haven't ever wanted to join.

I'm never going to feel obligated to identify them. 

Do you feel those images shouldn't be there?

They are (hopefully good-ish) examples of my work with amateur models.

Whereas uploading a Midjourney image of a person who does not exist would not be. It's a very clear line between the two.

Edited by Unfocussed Mike

I do - always, I consider it rude otherwise.  (Ditto models whose photos miraculously appear without a photographer present!)

But each to their own.

Edited by Gothic Image

Gothic Image said

Unfocussed Mike said

Theimagebear said

If you create images with no model or there is one but you choose to keep their identity private (a growing thing) then thats not what here is about.

You mean if I have a model who is not on purpleport and won't tell you who they are?

I have plenty of work on here that is demonstrative of what I can do as a photographer, that features three people who enjoy working with me but haven't ever wanted to join.

I'm never going to feel obligated to identify them. 

Do you feel those images shouldn't be there?

They are (hopefully good-ish) examples of my work with amateur models.

Whereas uploading a Midjourney image of a person who does not exist would not be. It's a very clear line between the two.

Edited by Unfocussed Mike

I do - always. Why wouldn't you?  (Ditto models whose photos miraculously appear without a photographer present!)

They get identified as "model not on purpleport" if that is what you mean. 

That's all you're getting because that's all I've told them I will say about them.

Not sure what the controversy is here? Who is it rude to? They aren't models. I'm expected to give their names to strangers on the internet?

Edited by Unfocussed Mike

Unfocussed Mike said

Gothic Image said

Unfocussed Mike said

Theimagebear said

If you create images with no model or there is one but you choose to keep their identity private (a growing thing) then thats not what here is about.

You mean if I have a model who is not on purpleport and won't tell you who they are?

I have plenty of work on here that is demonstrative of what I can do as a photographer, that features three people who enjoy working with me but haven't ever wanted to join.

I'm never going to feel obligated to identify them. 

Do you feel those images shouldn't be there?

They are (hopefully good-ish) examples of my work with amateur models.

Whereas uploading a Midjourney image of a person who does not exist would not be. It's a very clear line between the two.

Edited by Unfocussed Mike

I do - always. Why wouldn't you?  (Ditto models whose photos miraculously appear without a photographer present!)

They get identified as "model not on purpleport" if that is what you mean. 

That's all you're getting because that's all I've told them I will say about them.

Not sure what the controversy is here? Who is it rude to? They aren't models. I'm expected to give their names to strangers on the internet?

Edited by Unfocussed Mike

Which is absolutely fine in that particular case.

Gothic Image said

Unfocussed Mike said

Gothic Image said

Unfocussed Mike said

Theimagebear said

If you create images with no model or there is one but you choose to keep their identity private (a growing thing) then thats not what here is about.

You mean if I have a model who is not on purpleport and won't tell you who they are?

I have plenty of work on here that is demonstrative of what I can do as a photographer, that features three people who enjoy working with me but haven't ever wanted to join.

I'm never going to feel obligated to identify them. 

Do you feel those images shouldn't be there?

They are (hopefully good-ish) examples of my work with amateur models.

Whereas uploading a Midjourney image of a person who does not exist would not be. It's a very clear line between the two.

Edited by Unfocussed Mike

I do - always. Why wouldn't you?  (Ditto models whose photos miraculously appear without a photographer present!)

They get identified as "model not on purpleport" if that is what you mean. 

That's all you're getting because that's all I've told them I will say about them.

Not sure what the controversy is here? Who is it rude to? They aren't models. I'm expected to give their names to strangers on the internet?

Edited by Unfocussed Mike

Which is absolutely fine in that particular case.

Surely friends/relatives/partners is the default for unnamed people on Purpleport. Especially among new portfolios.  We're a bunch of amateurs.

I'm not going to name people or say anything about them unless I'm specifically asked to by the subjects themselves.

They aren't on purpleport. I shouldn't even need to put "model not on purpleport". I just do it to stop people asking in the comments. 

FWIW I'm sure there is work on my portfolio where the model has chosen not to show tags, too. Their call, not mine. Almost all my work is old. Most of it not very good. I don't care why they do or don't.

Edited by Unfocussed Mike

Bob @ Fatbloke said

Isn’t it rather like using a camera on anything other than “M”? Quite a bit of skill producing AI, like modern cameras? :-)

It is, if the camera walked, set up the scene, composed it, waited for the right weather conditions and subject, pushed the shutter, and then edited the photo, offering you 4 edits and you repeatedly picking which ones your like and it then repeating and refining the process until you are happy - all based on you telling it "photo of cat".

So yeah, not too far off "P" mode on a modern camera really :-p

Russ Freeman said

Bob @ Fatbloke said

Isn’t it rather like using a camera on anything other than “M”? Quite a bit of skill producing AI, like modern cameras? :-)

It is, if the camera walked, set up the scene, composed it, waited for the right weather conditions and subject, pushed the shutter, and then edited the photo, offering you 4 edits and you repeatedly picking which ones your like and it then repeating and refining the process until you are happy - all based on you telling it "photo of cat".

So yeah, not too far off "P" mode on a modern camera really :-p

It's more like making a pizza from scratch or ordering from Dominoes.

I'd love to see the term 'artificial intelligence' drop out of use. Computers only do what they are programmed to do. They don't think and they are not intelligent.

Photowallah said

I'd love to see the term 'artificial intelligence' drop out of use. Computers only do what they are programmed to do. They don't think and they are not intelligent.

The funny thing is that techniques in the "artificial intelligence" bracket tend to drop out of it as they become better understood as classifier methods. We no longer really define Prolog programs to be "artificial intelligence"; they are "expert systems" instead because they describe codified/stratified/testable knowledge.

But as a class, it is a reasonable term for describing a bunch of programming techniques that do not emerge from simple algorithms or "expert system" databases, that represent models of biological intelligence better than they represent linear programming or statistics.

In some ways, these tools think a lot like basic parts of our brain think.

The model data that Stable Diffusion runs on, for example, is about 4.3 gigabytes in size. But it was trained on 2.3 billion images. This alone should show you that the data set is the result of a process of reinforcement learning: it cannot be using two bytes to describe every image it was shown. It's not an expert system database. That data does not describe "knowledge" in any sense that we would key it in.

Stable Diffusion genuinely has learned to make new images from that dataset -- from the abstract and actually not immediately inspectable data it contains. It has learned to associate the text with the images and to generate new plausible responses from similar words. There's no program for cats or elf ears or anything like it; you can't look into that data and see a program for Pollock's style. And you couldn't trace through the code watching it draw a face step by step.

It's a data set that resulted from extraordinarily labour intensive training of a neural network. This is a lot closer to how our brains work than any conventional linear or even parallel computer program.

Edited by Unfocussed Mike

Photowallah said

I'd love to see the term 'artificial intelligence' drop out of use. Computers only do what they are programmed to do. They don't think and they are not intelligent.

This is not strictly true......  Machine Learning, and Artificial Intelligence differentiate themselves by actually being able to use what they "learn" to influence what they continue to learn.

Unlike many humans......eh?

Bob @ Fatbloke said

Photowallah said

I'd love to see the term 'artificial intelligence' drop out of use. Computers only do what they are programmed to do. They don't think and they are not intelligent.

This is not strictly true......  Machine Learning, and Artificial Intelligence differentiate themselves by actually being able to use what they "learn" to influence what they continue to learn.

Unlike many humans......eh?

Yes. "Machine Learning" is the more useful precise term here. Not least because it's better on grant applications.

I personally subscribe more to Steve Grand's theory that a general pursuit of "artificial intelligence" is largely meaningless when uncoupled from artificial instincts/urges/reward-seeking (like he demonstrated in his Creatures games). 

But as an enclosing "folder" in which systems that are designed to replace human instinct are placed, "Artificial Intelligence" is not the worst two word label.

Edited by Unfocussed Mike

Russ Freeman said

Bob @ Fatbloke said

Isn’t it rather like using a camera on anything other than “M”? Quite a bit of skill producing AI, like modern cameras? :-)

It is, if the camera walked, set up the scene, composed it, waited for the right weather conditions and subject, pushed the shutter, and then edited the photo, offering you 4 edits and you repeatedly picking which ones your like and it then repeating and refining the process until you are happy - all based on you telling it "photo of cat".

So yeah, not too far off "P" mode on a modern camera really :-p

All those items mentioned are readily available on the likes of YouTube, for "photographers" to learn mechanistically.....as in copy without understanding.

Some will never improve, (which is fine, if they don't have aspirations to), but my point is (and was) that evolution, be it in silicon or otherwise, is not a threat, but an opportunity?

Analogy.

"Photographer".

Buy Camera.

Hire Model

Set Auto

Press Shutter and shoot 500 images

Find one that looks half decent

Voila.

Machine Learning just "shoots" a few million images, and picks the one that looks like a photo?

Devil's Advocate.

Edited by Bob @ Fatbloke

Unfocussed Mike said

Bob @ Fatbloke said

Photowallah said

I'd love to see the term 'artificial intelligence' drop out of use. Computers only do what they are programmed to do. They don't think and they are not intelligent.

This is not strictly true......  Machine Learning, and Artificial Intelligence differentiate themselves by actually being able to use what they "learn" to influence what they continue to learn.

Unlike many humans......eh?

Yes. "Machine Learning" is the more useful precise term here. Not least because it's better on grant applications.

I personally subscribe more to Steve Grand's theory that a general pursuit of "artificial intelligence" is largely meaningless when uncoupled from artificial instincts/urges/reward-seeking (like he demonstrated in his Creatures games). 

But as an enclosing "folder" in which systems that are designed to replace human instinct are placed, "Artificial Intelligence" is not the worst two word label.

Edited by Unfocussed Mike

And what is striking about these systems is that you can in no way tweak the code to improve output-specific behaviours in a system. You can't go in and, say, change a few instructions that affect brush strokes or colour choice or circles. Because they don't exist; they weren't ever written. 

These systems can only be improved by either changing their "neurology" and retraining the system, or at best by providing more specific training data sets to an existing system.

The code of the system is not doing the job of making pictures from words. The code of the system -- the bit the machine was actually programmed to do -- is the supporting mechanics of the engine that has learned to do that. The internal state of that running system is on two levels -- the interface data of the neural mechanics (parsed words and RGB image data) and the behaviour state of the neural mechanics (connections, connection weights and point values, I suppose).

And as a consequence, these systems contain genuine unexamined biases, quirks, and misunderstandings that may take months to find.

This is a lot closer to natural intelligence than algorithmic behaviour.

Edited about a squillion times to make it make better sense.

Edited by Unfocussed Mike

Bob @ Fatbloke said

Machine Learning just "shoots" a few million images, and picks the one that looks like a photo?

Devil's Advocate.

Edited by Bob @ Fatbloke

The astonishing thing about Stable Diffusion -- if I understand it and I am not sure I do -- is that it seems to have been trained on something inbetween.

It models the opinions of human viewers. 

So it's not that it picks the ones that look like a photo. It effectively picks the ones that its training has led it to believe human viewers will consider a good (high-scoring) visual illustration of the words in the description.

The more I read about this, the more freakishly sci-fi it becomes.

Edited by Unfocussed Mike

I work with AI/ML (not in a photography sense here.....the two just don't overlap in my world).

Banning it (other than from competition space), just seems like a rather Luddite approach?

Why not ban anything other than Manual cameras.

Anything not SOOC

Photoshop (with its AI filters and functions)......Herecy.

It is just another step on the road to the future....surely?

Bob @ Fatbloke said

Banning it (other than from competition space), just seems like a rather Luddite approach?

Good examples of why other communities have felt the need to do it here:

https://waxy.org/2022/09/online-art-communities-begin-banning-ai-generated-images/