Midjourney?

[bjorke]

Midjourney's been trained on the same stuff people have been flooding onto the internet, and it regurgitates material much like what it's seen. Which is pretty much what photographers do, no? Like (imho good) photography, Midjourney images are very much of their time, and they plumb new sinkers through our media-soaked environment. It doesn't seem that far, for me, from Robert Heinecken repurposing ads, or William Klein's Broadway at Night, or Fox-Talbot scraping ideas out of painting for his selfies. Like open-air photography, using generative AI databases is about spelunking the collective visual culture. You just need to ask the magic box exactly the right question.

 

[MattKing]

Hey - a post from bjorke!
Good to see you post again.

 

[Don_ih]

You just need to ask the magic box exactly the right question.

Is that exactly the right question to get it to produce the realization of your idea or is that question the right question when you recognize the merit of the result?

If you take a landscape photographer as an example, the photo is a particular view that has been more or less created by the photographer through careful selection of several (or all) of viewpoint, framing, exposure, developing, enlarging - as activities the photographer alone directs. His or her activity is influences by knowledge and experience - including that of other photos. other scenes, other photographers, other art forms, etc. Those influences are actually so numerous and potentially so subtle that they would be practically impossible to fully outline. But that does not make the photo a product of those influences.

The AI generated images, on the other hand, are the product of what could be considered influences - as elements gathered by the cataloguing AI "memory". So, when an object is proposed, it glues together aspects of related (perhaps identical) objects it has catalogued into a new object to fulfill the request.

I think the end result is more like picking the off-the-rack shirt you like than being anything like a tailor. The decisions regarding design have already been chosen for you.

 

[bjorke]

Landscape photography is a reaction — to the location, available materials, weather, and informed by a considerable catalog of existing mannerisms, ones that define a photo as “good.” The photographer has limited agency in choosing. So too a generative model. You choose some things, exploit some things, click a button, wait.

 

[Bill Burk]

Interestingly this thread started when people were just becoming familiar with the images.

Already there has been enough backlash that the novelty has worn off, everyone universally hates it and can spot it a mile away.

Thought I read an art show or craft fair application explicitly forbidding artificial intelligence based images.

 

[bjorke]

Funny that “everyone universally hates it” yet Midjourney has become one of the world’s largest cloud GPU consumers, with all of the consumption paid for directly by users (not indirectly a la Facebook or Google).

Despite hand-wringing on text forums, “everyone” seems to be voting with their credit cards.

 

[Don_ih]

Landscape photography is a reaction

I didn't imply it wasn't a reaction. The extent to which it is a reaction would be difficult to assess. And, from a purely deterministic point of view, there is no discernible agency whatsoever. But the magic of being human is the ability to deny any and all those claims and act as though there is agency and the ability to both choose and create.

Already there has been enough backlash that the novelty has worn off, everyone universally hates it and can spot it a mile away.

The novelty has far from worn off. The use of these images has not even taken a toehold in the world yet. Within 20 years, there will be so many AI-composed images and they will be impossible to distinguish from any other images. If you are in a position that needs imagery for ad campaigns or anything similar, imagine the appeal of just typing in a request and immediately getting back 10 options that are all "good enough" - so much cheaper and less bothersome than getting people to do it. Photography that doesn't need to pay a photographer or a model, doesn't need to rent a space.

 

[bjorke]

 

[warden]

For those not yet tired of AI here’s a fascinating article. https://www.theguardian.com/technol...-of-ai-artificial-intelligence-dall-e-chatgpt

 

[Sean]

I've been playing around with the new version that just came out.

"street photo Tokyo"

Cyanotype

street photo mural of robots

"an elegantly designed pair of loudspeakers"

There isn't much it can't do if directed properly (far better than my directions). And this is very early days, so imagine Version 10.0 which will be orders of magnitude more competent.

It can make you ask yourself, what will the point of real photography be if anyone can conjure up any photo you desire within seconds? But it didn't take me long at all to get bored of it, and well, I am nearly 50yrs old and still not bored of actual photography. I like to get out there, me, my eyes, and a camera, explore and capture my own reality. The novelty of 'prompting' fades quick.

The only thing I find interesting or fun about Midjourney is putting in some unusual prompts and seeing what happens.

 

[Sean]

recycling a prompt multiple times generates a new content each time. In this case, "a robot made out of old camera parts"

 

[awty]

recycling a prompt multiple times generates a new content each time. In this case, "a robot made out of old camera parts"


View attachment 332885


View attachment 332886


View attachment 332887


View attachment 332888


View attachment 332889


View attachment 332890

So how much info do you need to give it? Did you just ask for camera robots and that's what it gave you, or can you manipulate the outcome more?
That would save a lot of work, I have to spend hours/days making stuff by hand.

 

[Sean]

So how much info do you need to give it? Did you just ask for camera robots and that's what it gave you, or can you manipulate the outcome more?
That would save a lot of work, I have to spend hours/days making stuff by hand.

All I asked for was "a robot made out of old camera parts" and it generated the above results. You can definitely steer it. For example on this one, I asked for "a wide angle view photograph of dirty damaged robot made out of old camera parts, standing on a filthy dystopian city street, depressing lighting and this was the result about 45 seconds later:

Or another variation:

"a photograph of a colorful painted robot made out of old camera parts, standing in Rome"

 

[Sean]

The last robot, I told it to make more variations of it:

And if you liked one of the variations, you can make variations of that one and so on.

 

[Sean]

you can put in random prompts just to see what happens, it takes around 45 seconds for the result

"a collection of futuristic sneaker designs "

"an assortment of futuristic and unique 3D printed home designs"

 

[retina_restoration]

These AI "creation" tools are all trained on imagery parasitized from the internet - from Getty Images, from magazines, from Flickr and Instagram. You might want to read this. There are class action lawsuits in progress, suing for misappropriation and copyright infringement. If an AI tool like Midjourney uses a dataset trained on YOUR photographs, its a violation of your IP rights.

 

[warden]

"a collection of futuristic sneaker designs "

View attachment 333102

The shoe designs are fun, and I can see a path to commercializing them once developed in 3D CAD space. This is a good example of using AI to develop a wealth of idea seeds, and then using the output as a jumping-off point for an actual designer to create a manufacturable product. It's amazing.

 

[Helge]

If we pull back to a birds eye view for a second, I do not see how this can not in some way and to some degree help film photography?
Negatives and slide is very hard to fake perfectly, and kind of pointless too unless we are talking a court case or similar.

That’s one of the aspects of film photography that raises the greatest amount of ire, when you raise the point to aggressively digital people.

But it’s basically and viscerally true.
Film is just that much more authentic.
You have something physical to show as proof.

 

[Arthurwg]

I realized that we were in hell when that Everything film won the Oscars.

 

[awty]

Everything is so formulated it may as well be done by machines.

 

[Sean]

These AI "creation" tools are all trained on imagery parasitized from the internet - from Getty Images, from magazines, from Flickr and Instagram. You might want to read this. There are class action lawsuits in progress, suing for misappropriation and copyright infringement. If an AI tool like Midjourney uses a dataset trained on YOUR photographs, its a violation of your IP rights.

I have seen the arguments and they hold a lot of water. I am hugely concerned over these search engines ripping our data here and giving people direct answers via our content, leaving them no need to visit. In my view, this is a type of copyright theft. I see many lawsuits coming and I hope that websites will not be fed into these AIs by default and you must opt-in.

The artwork is tricky as they will no doubt fire up some mental gymnastics and claim that human artists do the exact same thing the AI is doing.

 

[bjorke]

Everything is so formulated it may as well be done by machines.

As I recall, a camera is a machine, and photographers have a limited range of methods for altering the images generated from it.

Also, almost everyone uses the tool in a rather banal way, in the all-pictures-are-nouns way: "cat in spacesuit with rings planet in the sky"

This has two problems: (1) no one can say that doesn't really look like a cat in a spacesuit, because no one has ever seen one. This problem exists for lots of fantasy media. (2) there's no reason for a cat to be in a spacesuit. No verb.

Here is a proposed image of myself posing for a portrait, aged 112:

Here's an image about dealing with complaints about AI imagery:

all good pictures have a verb. A "decisive moment" some call it. This was true long before photography, and after it.

 

[film_man]

Think of it like this:

Industrial production of plates and other "ceramics" has basically put out of work everyone not targeting the higher end "artistan" or whatever market. Midjourney will basically put out of work all the mundane stock photographers who spend half a day doing photos and trying to sell them for 5c on Alamy or shutterstock.

On the other hand, said industrial plate production has had no effect on the hobyyist ceramic market, with local schools and weekend workshops for people that just want to have some fun being unaffected.

Reportage/documentary photographers need not worry, at least not until we end up in a Minority Report future. I'm guessing wedding photographers are safe too, I'm sure some morons will use some phone pics to run through midjourney to create photos from a wedding they didn't have but those people wouldn't have paid for a decent photographer anyway. Portrait photographers..probably safe too, I mean as much as I like the funny avatars I made for myself, which interestingly all look like a hermaphrodite elf from Mordor.

Now..commercial photographers...ahh.."computer get me an image for the next Vogue cover with <insert name> wearing the latest clothes from <insert brand>." How long before we see that I wonder. In fact..."computer just make the next issue of Vogue. Have models a/b/c and clothes x/y/z. Rehash the same crap for April as we did 5 years ago".

 

[bjorke]

did this fully-synthetic image with colleagues in 2001 for MAXIM Special Edition Cover - over twenty years ago!

People whined about it but c'mon, the difference in likelihood of the average Maxim reader dating Cameron Diaz (previous cover) vs Aki Ross (this character) were essentially the same. AND: this photo went straight to the page, no retouching.

 

[awty]

As I recall, a camera is a machine, and photographers have a limited range of methods for altering the images generated from it.

Also, almost everyone uses the tool in a rather banal way, in the all-pictures-are-nouns way: "cat in spacesuit with rings planet in the sky"

This has two problems: (1) no one can say that doesn't really look like a cat in a spacesuit, because no one has ever seen one. This problem exists for lots of fantasy media. (2) there's no reason for a cat to be in a spacesuit. No verb.

Here is a proposed image of myself posing for a portrait, aged 112:

View attachment 333270

Here's an image about dealing with complaints about AI imagery:

View attachment 333271

all good pictures have a verb. A "decisive moment" some call it. This was true long before photography, and after it.

A camera is a box with a hole up one end and something to catch the light up the other.
We see what we're taught, just like the program.