Near infrared by DALL-E 2

[Stefano]

This is a very informal topic, but I still find it interesting because it shows how deep learning models such as DALL-E 2 have "learned" what a NIR photo looks like (especially the Wood effect).

 

I tried the same with UV, but only got UVIVF images, which shows how little known reflected UV photography is. These AI models are a reflection of what people know (and what is posted online). Not surprisingly, NIR photography is much more known than the UV counterpart.

 

All images have been entirely generated using DALL-E 2. No edits were done, posted at full resolution (1024x1024). The prompt for all images is "Near infrared photography".

 

 

 

 

 

 

 

 

 

[Andrea B.]

Strange that these AI images seem so familiar to me. The AI must have learned IR from Kolari or Life Pixel pages . 

[nfoto]

AI-generated IR kitsch. Good grief. There is no hope for the future?

[Stefano]

Well, if you want to take photos of actual subjects (real people, real places, etc.) photography will remain the only way, hopefully.

 

Artistic and abstract images are a different story. Many artists have criticised AI because it could steal their job, or because it was trained on images made by real artists without giving credit.

 

I don't think photography will ever die, just like painting didn't die with the birth of photography, like many painters feared back then (although the importance of painting for recording the real world has faded).

[nfoto]

Running a mediocre image through "creative" PS-filters won't make it less kitsch.

[Stefano]

Also, this is for Birna: is there any "real" plant in those images? Or plants that could be loosely identified?

[Fandyus]

I find it funny how it goes out of its way to add hotspots. I guess most IR images out there are taken by people who don't care too much.

[Fandyus]

6 hours ago, Stefano said:

Also, this is for Birna: is there any "real" plant in those images? Or plants that could be loosely identified?

I am not sure Birna will reply but I would think so. The AI is really good at pattern seeking so it can probably pick out a pattern where certain trees look like oaks for example, so it adds a tree that looks like an oak. But it could also just combine random features of different species to make something unintelligible.

[nfoto]

Some of the tree silhouettes have a faint similarity to *known* species. However, to the best of my knowledge there is no tree able to exhibit a levitating behaviour or having disjunct growing trunks....

 

The uncanny similarity to an android going amok  with the smudge tool in PS is more than I can stomach at present.

[Stefano]

Thanks.

[Fedia]

The first IR images I took with an uncoverted canon 600D and a 720nm filter looked just like this, just it had cows in them.

 

I think the way these IA images reveal the mediocrity of what is usualy donne in IR is a good critical tool for us. It invites us to do things differently. To me it's always interesting to see what IA come up with. It's uncanny aspect only strikes us because of what it reveals on ourselves.

[dabateman]

So this is what the training sets might look like. Or what the bots could find freely available on the internet. 

So time for something completely different. 

 

[colinbm]

If this is what AI is about, I think it has failed !

 

[lukaszgryglicki]

I'm a programmer with 20 years of experience and I don't believe in "AI" - there is no such thing.

 

[Stefano]

I think AI has pros and cons like any new technology. It can help decipher old languages, or help diagnose cancer (at least as an aid, I still trust real doctors more). Of course it can also be used with malicious intent, like generating deepfakes of public figures (which is honestly scary).

 

There are many areas where humans perform better than AI, at least for now. Surely Lukas writes better codes than ChatGPT, but maybe AIs could still be helpful for certain tasks.

 

AI-enhanced photos is something I really don't like. I hate it when my phone visibly oversharpens my photos, at least now most phones offer a "pro" mode that gives some control to how the camera takes photos. So AI-processed images are even worse. I prefer to capture reality rather than capture a nice photo, I think this is what photography is about (although some creativity is still allowed, like lenses with square bokeh).

 

I mainly used DALL-E 2 to generate abstract images, like thinks I saw in my dreams or places I imagined as a child. But I wouldn't use it for improving real photos or images that depict reality.

[lukaszgryglicki]

I think the name is totally misleading - there is no "intelligence" in any computer. This is just a brutal force of learning nets and algorithms - any intelligence anywhere in computers is in the authors of the source code of algorithms.

I agree with all you said, those algorithms can be useful (and are) in many places, but I will never agree that there is something like "artificial intelligence" in the machine.

 

[Stefano]

Yes, I agree that intelligence is something a machine doesn't have.

 

These generative models do show some remarkable behaviour that resembles intelligence (generating images from text still sounds sci-fi to me, sometimes I really say "wow, a machine created this image!"), but it's not true intelligence.

 

I don't know much about neural networks, but I think they imitate the brain, having "neurons" connected together which fire or not depending on the inputs they receive. But being a machine its actions are deterministic: given the same input, the network will give the same output. That's where I think it differs from an actual brain.

[photoni]

I've been frequenting Flickr for 18 years, at the beginning everything was new, the transition from film to digital made us discover and rediscover photography, as an immediate memory, like a Polaroid, giving us the possibility of making free multiples and being able to share it.

Then I used a Hasselblad with digital back for work, and a compact to photograph my young children and the holidays; I shared some photos with everyone, even those of my ancestors from 100 years ago with near and far relatives from Brazil

 

Going back to today everything is becoming more and more artificial and false, in addition to the exasperated retouching of images, now there is the possibility of recreating non-existent people.

The group "More Human than Human: Human Artists creating AI Art" was born 4 months ago, it already has 6,500 images, some indistinguishable from reality.

 

It scares me to read what the author of this picture "Lost in The Crowd" writes

 

The problem is not the use of AI but the abuse.
that's why actors and screenwriters are striking in the USA.

 

I've always read a lot of science fiction books, but I didn't think about such a fast exponential growth, and about the "degeneration" of the image with AI.

 

Now I'm happy to be retired, I no longer have to worry about working for three days to create an image that a machine "emulates" in 3 seconds.

 

In addition to the PRO equipment that I no longer use, and the Full Spectrum and the iPhone...
I continue to take pictures with BW film, I develop and print everything in house.

I can use ecological developments such as the one with coffee and vitamin C,

I take the fixing that contains silver to the ecological center.
I print cyanotypes that use poor elements such as iron sulphate.

 

AI exists... but ...  F**k

[StephanN]

Remember this movie? https://en.wikipedia.org/wiki/The_Terminator At that time, everybody was laughing at the ridiculous plot, but one friend of mine simply said "Just you wait". Well, I guess, nobody's laughing now.

[Fandyus]

On 7/20/2023 at 2:30 AM, dabateman said:

So this is what the training sets might look like. Or what the bots could find freely available on the internet. 

So time for something completely different. 

 

Haha, that's what I've been trying to do with my Aerochrome emulations and such. I've also recently used a 510nm longpass stacked with a grb3 equivalent by Tangsinuo and it makes everything look like a mix of Lomochrome Purple and normal visible color photos.

[Fandyus]

23 hours ago, Stefano said:

I think AI has pros and cons like any new technology. It can help decipher old languages, or help diagnose cancer (at least as an aid, I still trust real doctors more). Of course it can also be used with malicious intent, like generating deepfakes of public figures (which is honestly scary).

 

There are many areas where humans perform better than AI, at least for now. Surely Lukas writes better codes than ChatGPT, but maybe AIs could still be helpful for certain tasks.

 

AI-enhanced photos is something I really don't like. I hate it when my phone visibly oversharpens my photos, at least now most phones offer a "pro" mode that gives some control to how the camera takes photos. So AI-processed images are even worse. I prefer to capture reality rather than capture a nice photo, I think this is what photography is about (although some creativity is still allowed, like lenses with square bokeh).

 

I mainly used DALL-E 2 to generate abstract images, like thinks I saw in my dreams or places I imagined as a child. But I wouldn't use it for improving real photos or images that depict reality.

Oversharpened phone photos are not AI. If you want to see what real AI sharpening does, try Topaz software. I own the Denoise AI and even it has some rather impressive sharpening capabilities that can make you pictures significantly cleaner without introducing ugly halos like the simple formula used by smartphones would.

The way smartphones butcher their raw data is honestly painful to me, being a guy who really cares about ACTUAL image quality. But the average person doesn't care, so I don't expect this to ever improve. It's like a race to the bottom in who can make their photos look like overprocessed digital garbage the most.

[Fandyus]

As per my take on AI. I can understand why you guys dismiss it or outright dislike it, I'm well aware it's been demotivating artists around the world from actually picking up their drawing tablets and such.

At the same time, I also realize that this is the natural progress of computers and unless you don't want us to ever have more advanced sci-fi technology, you can't expect this not to happen. Computers will eventually be capable of faithfully mimicking human behavior and I do not know what the implications of that will be, but everyone knew it would happen someday. It might bring destruction, it might bring us more comfortable lives where we only have to work 3 hours a day instead of 8. For my own sake, I don't want to think about which one is more likely.

Also, the abundance of cheap AI art will most likely make the value of physical media skyrocket as it will once again be the only way to enjoy art that is 100% authentically human, so pick up your graphite and oil paint I suppose. Or get some film, if you can even afford the development and scanning, I can't

[Stefano]

7 hours ago, Fandyus said:

Oversharpened phone photos are not AI. If you want to see what real AI sharpening does, try Topaz software. I own the Denoise AI and even it has some rather impressive sharpening capabilities that can make you pictures significantly cleaner without introducing ugly halos like the simple formula used by smartphones would.

The way smartphones butcher their raw data is honestly painful to me, being a guy who really cares about ACTUAL image quality. But the average person doesn't care, so I don't expect this to ever improve. It's like a race to the bottom in who can make their photos look like overprocessed digital garbage the most.

Yes, you are right. What I meant is that I don't like the normal oversharpening, so I like AI-enhancing even less.

[Fandyus]

4 hours ago, Stefano said:

Yes, you are right. What I meant is that I don't like the normal oversharpening, so I like AI-enhancing even less.

May I ask why? Actual AI sharpening is semi-aware of what is it that is being sharpened and so it can keep the various unpleasant artifacts at bay (for the most part), that is the strength of neural networks.

[Stefano]

AI might be aware of what it's doing, but it's still not the "real world", it's not the raw output of the lens on the sensor. It is something that comes from outside, in a way. Even though I get your point, AI might make the right choices when improving an image.