Is there really a difference between Photoshop and AI?

Wondrous everyone is allowed an opinion it’s a forum. and everyone’s art is indevidual why would you make a snipe comment like that ? Did I attack your work ? I was only offering my point of view that everyone is entitled to create however they want to.

Retouchguy I was only offering my point of view , it's a forum. Everyone is entitled to think what makes art and whether it fits their definition of art.

That’s what I’ve said since the start of the conversation what you just said was “some people clearly don’t know what art is “ so that’s the opposite of everyone is entitled to there own art that’s an attack on my art which is pretty low

Edited by Retouchguy

Tom Cruise said “Don’t think. Just do” in Top Gun Maverick (although I believe he may have borrowed it from a guy called Horace who said something similar a couple of thousand years earlier).

A certain degree of mindfulness can be a good thing.

I try not to overthink stuff. I try to be Zen. *

AI is a tool. Use it, or not. The results are what count, not the process**.

*I’ve recently switched to doing most of my editing in Lightroom on an iPad mini (from Photoshop on a PC with a big screen). Very different experience, much more mindfulness. Adobe call it “AI masking”. I prefer to think there’s some magic fairy dust involved.   Same thing.

**although I admit to shooting a bit of 5x4 film, which is pretty much the opposite to what I just said. Ideally, I’d like to practice enough to do that without conscious thought.

So….   Just do whatever makes life fun.

I’m a bit dubious of the phase “creating images by text” as it kind of assumes your the artist when in reality when Pedro Picasso types “woman on beach” and gets something back far above his skill level, a composition he doesn’t understand etc, lighting he never dreamt of, he’s probably more like commissioning midjourney to create all the art with a huge amount of autonomy , then he’s allowed to use it under license, however due to the disruptive nature of ai, the laws on copyright, licensing, etc are hard/impossible to apply at this stage in time. The work is also not produced via any lens based technology, ie it doesn’t fall under the banner of photography. Ps older tools where pushing edited images towards digital art but It didn’t really have the  “ask a much better artist than yourself to help you out” buttons which are now availavle. Ai imaging tools are closer to using fiver.com and claiming the results your creativity.

I was thinking about this a lot over the weekend and thinking back to the kinds of shots I used to do for bands promos and album covers. Specifically, how at the time, there were no AI assistants on hand.

In so far as "I ask ChatGPT to make me a prompt and then put it into an image creator" that's a no brainer. It's not 'art' any more than creating a good Excel spreadsheet is. It's a skill.

But back to the point, I did this one and went back to my old ways of not taking any shortcuts or plug-ins and just layering and blending until I got the effect I wanted.

But when I got the end of it, my first thought was "This probably looks like AI". Not maybe the newest versions, but definitely has that vibe. 

Point being, I have noticed the trend for processing is getting very much "less is more" and back to the quality of the source image.

It feels like Photographers might be in a bit of an arms race for realism with the machines. Which might be a good thing in the shorter term.

MacMaghnuis said

I love editing images in Photoshop. All the images on my portfolio have been photoshopped as anyone with an eye can tell. I've had criticism that I take it too far and that some of the images are veering into AI and lack character or personality. That is is the flaws in a person's face that are important. Does the spot healing tool or the patch tool or the liquify filter count as AI these days? Is there a creative difference between photoshopping and image and creating an image using text?

There is "AI" and there are "smart" algorithms branded as AI .

Spot healing, content-aware fill, content-aware stretch are all 'smart' to a degree, but you decide what should be filled / healed / patched.  and the software does it the best way that programmers have worked out.  I removed some lint from what a model was wearing with the dust and scratches tool and that doesn't feel like crossing a line which the program recognizing spots would. Liquify and puppet warp, are human-controlled (but smart) algorithms... all of these start with a photograph and process it to make it more what is in its creator's mind's eye. 

"Cybergraphs" as I've taken to calling images which are created without light hitting something light sensitive, can certainly be a way for people to be creative. Not the same as a photograph, sometimes interchangeable with photographs. And there are blurry edges - generative fill vs content aware fill for example, where AI is generating a small picture and blending it into another one.  

Perception said

I’m a bit dubious of the phase “creating images by text” as it kind of assumes your the artist when in reality when Pedro Picasso types “woman on beach” and gets something back far above his skill level, a composition he doesn’t understand etc, lighting he never dreamt of, he’s probably more like commissioning midjourney to create all the art with a huge amount of autonomy , then he’s allowed to use it under license, however due to the disruptive nature of ai, the laws on copyright, licensing, etc are hard/impossible to apply at this stage in time. The work is also not produced via any lens based technology, ie it doesn’t fall under the banner of photography. Ps older tools where pushing edited images towards digital art but It didn’t really have the  “ask a much better artist than yourself to help you out” buttons which are now availavle. Ai imaging tools are closer to using fiver.com and claiming the results your creativity.

This is a very interesting take, and I agree.

Except that I think that what is going on is the commissioning of a much _worse_ artist, in the sense of actually creating art.

Midjourney isn't an artist -- it's an eager-to-please, focus-group-trained robot that has been shown all art and experienced none of it.

Every single terrible BLBP image on this site and others, every swinger meetup photo on Flickr, every colour-popped black and white photograph of the only flower that grew in the garden this year posted to someone's facebook friends, every failed GCSE photography portfolio image is more art than Midjourney can ever produce.

The terrible gap here is a Dunning-Kruger gap: people who as you say do not understand, who do not know that their transition to midjourney artist from bad photographer is glaringly obviously inauthentic to people viewing the work.

Friends don't let friends use generative AI when they could be out there taking bad photos instead.

Edited by Unfocussed Mike

CalmNudes said

There is "AI" and there are "smart" algorithms branded as AI .

Spot healing, content-aware fill, content-aware stretch are all 'smart' to a degree, but you decide what should be filled / healed / patched.  and the software does it the best way that programmers have worked out.  I removed some lint from what a model was wearing with the dust and scratches tool and that doesn't feel like crossing a line which the program recognizing spots would. Liquify and puppet warp, are human-controlled (but smart) algorithms... all of these start with a photograph and process it to make it more what is in its creator's mind's eye. 

"Cybergraphs" as I've taken to calling images which are created without light hitting something light sensitive, can certainly be a way for people to be creative. Not the same as a photograph, sometimes interchangeable with photographs. And there are blurry edges - generative fill vs content aware fill for example, where AI is generating a small picture and blending it into another one.  

Right. 

I mean I think it doesn't help that the broad history of the "AI" discipline is that technologies start out being called artificially intelligent but end up being called statistical methods or knowledge bases; almost all AI technologies start out as a belief in a step towards AGI and end up Poplog or Python libraries. Is a small back-propagating neural network classifier "AI" still? Perhaps not, because programmers understand it well enough to operate it on roughly the same level as they operate Naïve Bayes classifiers.

So there are plenty of things that have been called AI in the past that we would not call AI now; the history of machine vision is littered with these things.

But at the same time, there's a hot-topic bad faith argument that "simple" algorithmic things like content aware fill or the healing brush, or even the deep convolutional neural network upscalers and denoisers are fundamentally the same as generative tools, as if to say "oh what are you complaining about, you've been using these things all along".

Everyone who pushes this argument knows it is fundamentally untrue. To me it's a mark of intellectual dishonesty every time I see it; I lose respect in people.

Edited by Unfocussed Mike

Unfocussed Mike said

CalmNudes said

There is "AI" and there are "smart" algorithms branded as AI .

Spot healing, content-aware fill, content-aware stretch are all 'smart' to a degree, but you decide what should be filled / healed / patched.  and the software does it the best way that programmers have worked out.  I removed some lint from what a model was wearing with the dust and scratches tool and that doesn't feel like crossing a line which the program recognizing spots would. Liquify and puppet warp, are human-controlled (but smart) algorithms... all of these start with a photograph and process it to make it more what is in its creator's mind's eye. 

"Cybergraphs" as I've taken to calling images which are created without light hitting something light sensitive, can certainly be a way for people to be creative. Not the same as a photograph, sometimes interchangeable with photographs. And there are blurry edges - generative fill vs content aware fill for example, where AI is generating a small picture and blending it into another one.  

Right. 

I mean I think it doesn't help that the broad history of the "AI" discipline is that technologies start out being called artificially intelligent but end up being called statistical methods or knowledge bases; almost all AI technologies start out as a belief in a step towards AGI and end up Poplog or Python libraries. Is a small back-propagating neural network classifier "AI" still? Perhaps not, because programmers understand it well enough to operate it on roughly the same level as they operate Naïve Bayes classifiers.

So there are plenty of things that have been called AI in the past that we would not call AI now; the history of machine vision is littered with these things.

But at the same time, there's a hot-topic bad faith argument that "simple" algorithmic things like content aware fill or the healing brush, or even the deep convolutional neural network upscalers and denoisers are fundamentally the same as generative tools, as if to say "oh what are you complaining about, you've been using these things all along".

Everyone who pushes this argument knows it is fundamentally untrue. To me it's a mark of intellectual dishonesty every time I see it; I lose respect in people.

Edited by Unfocussed Mike

Indeed. Voice control, handwriting recognition, playing chess. All difficult problems which were once seen as needing degrees of machine intelligence we'd never achieve. 

For the bad faith thing... There are people who view anything other than an out-of-camera JPEG, with the camera set to factory defaults as cheating* ( and had the same view of any dark room work beyond developing with the recommended concentrations, temperatures and times and printing with equal exposure time to the whole print, on a standard contrast grade of paper) 

But people extrapolating from one kind of "OK" or one kind of "not OK" to everything is always bad (Occam's Razor - don't assume deliberate dishonesty, when a lack of comprehension may be be the cause). A bit of generative fill is photoshop is not the same of creating a "Cybergraph". Nor is creating a (photorealistic) image by machine intrinsically bad - it's a different kind of a picture with it's own pros and cons. 

* And lord knows what they make of phone cameras which will take half a dozen shots and montage something together where each person in the group looks their best - i.e. a a straight from the camera fake....  The room where I'm sitting has half a dozen panoramas hanging, so it's OK to join images ... but what about one I did recently where I put the best halves of two near identical shots together to produce a scene which was never in front of the camera. 

Think the issue is similar to painting yourself uninstructed other than your natural method of painting, than painting by numbers than painted by another , however in AI the other is in a sense at best a forgery of skill.

CalmNudes said

Indeed. Voice control, handwriting recognition, playing chess. All difficult problems which were once seen as needing degrees of machine intelligence we'd never achieve. 

Playing chess is kind of the extreme example of the "is it still AI?" puzzle isn't it?

On my uni course err three decades ago now, I recall chess being a sort of grand-challenge problem, and for sure many of the techniques in chess algorithms are kind of basic, core techniques in AI, used in all sorts of decision tree management, problem-space management problems; it's still foundational to AI.

But do people think of a chess-playing machine as "AI" in the way they are now being sold AI? I think even the colloquial understanding of chess-playing computers is now that they aren't clever so much as efficient in searching for the best moves. Colloquially, people understand that chess computers "look ahead", ruthlessly. Whereas experienced Go players would perhaps be inclined to think of computer Go players as artificially intelligent still.

One can place, say, the healing brush or content aware fill somewhere in that continuum. PatchMatch is more like an equivalent of one of those fundamental algorithmic building blocks (say something naïve like minmax).

Part of this, for sure, is my own bias in preferring to think of AI as the more instinctual end of the spectrum (neural networks) rather than expert systems. And even approaching this bias isn't new -- I remember being criticised for it even thirty years ago. The expert systems guys back then were quite ready to defend their work both as artificial intelligence, and as a necessary functional stage for neural networks to sit on top of. I was gleefully shredded for my view by a professor at the time!

But I can't help thinking he'd be withering about generative AI, were he still with us (he too is gone almost thirty years now). I have said elsewhere that I think LLMs and GANs are the empty calories of AI research; they will make research groups fat (cash rich) and stupid (focussing on the easy, worthless papers).

Edited by Unfocussed Mike