AI search: for instance “bird with red feather”, “group photos”, “wine”, “couple selfie in sunset”
On samsung galaxy S phones (i have S23+), this kind of search is really good. Recently, I looked for a “Crozes” and it found 2 pics of bottles, a pic of blackboard at a restaurant with the word on it, and a paper invoice with it !!
I submit that advanced technology such as recommendation algorithms and lots of machine learning is already degrading photography and human life. Of course, there may be some exceptional use cases such as AF, and I already dealt with that topic in a previous post.
As for aggressive retouching, I’m already well-aware of that. But what aggressive retouching did not have in the past is:
The tools that were used for aggressive retouching never had the potential to grow in complexity to the point that they would require gigawatts of power to train and gigabytes to store.
Aggressive retouching was difficult and not widespread or quick, and therefore were not seductive to the general photographer
Agressive retouching methods did not contribute to a large sphere of technology (ML/AI in general) that as a whole is detrimental to humanity.
I am not speaking here of some kind of apocalypse or terminator scenario, but the eventual forcing of features (or components) that are huge and also the contributing to the degradation of a healthy atmosphere in which to share photograpy.
If you are the type of person who only cares about having a handy tool for yourself and you have a lot of space or bandwidth to use it, my arguments won’t affect you. But I do care about the more distant future and I think that is what’s in peril here.
Your views on this topic are by now well documented in this forum and remind me of the complaints artists had about photography in the 19th Century.
I repeat what I said earlier. What may be good for you may not be good for me. Darktable is a tool for all of us, whose utility is increased by the free labor of those who contribute code to its development. We should evaluate future contributions based on the merit of what is developed, not what is feared.
These are very broad claims, and I don’t see any support for them, especially about human life. But I guess photography is doing just fine too; or if it isn’t by some metric, it is hard to attribute that to ML.
Sorry, I don’t understand what you are talking about. Almost all Darktable modules are optional (except for those like output/input profile). Even if someone made a module with LLM-based denoising or improved mask selection, no one would be forced to use it. Just like no one is forced to use any other module.
Mass surveillance is heavily reliant on these very same algorithms and it’s currently being deployed everywhere. There are big lobbies (Palantir) creating a market where there is none, or really the need for one, just to sell this tech to our governments. it has already infected most of Europe and it’s only gonna get worse and more players get into the game.
The most curious thing is that governments don’t even have a list of requirements before acquiring these services, the requirements are being developed and created by these lobbies to then serve this tech to them.
Recently in Texas there has been a lot of commotion due to massive installation of surveillance cameras to track you and your car everywhere, under the pretense of securing your vehicle from “robbers”.
Of course this is only the negative side. It’s really hard to say if it’s a net negative overall for humanity. We can now do faster MRIs using similar technologies, improving patient throughput, how many lives are being saved or extended due to not having to wait so much for such an exam?
Don’t forget the influence of public (voters’) opinion… Some political parties are rather hammering on the “growing insecurity”. Surveillance (cameras) are an easy answer to those concerns… Not installing them means the opposition (extreme right, quite often) gains seats next election. They will install them with any second thoughts :" See, we are actually fighting the insecurity" (and they get the money by cutting subventions used to help the poor).
Sorry, I still don’t understand how these are the “very same algorithms” that would be considered for inclusion into Darktable (even hypothetically).
ML is a very wide and varied field, and even narrower subcategories like neural networks have a lot of diverse algorithms, implementations, topology setups, etc.
Sure, I understand how the label “AI” is being used to sell everything, and a couple of years ago if a company had even a remote claim to using “AI” in a product or service, investors would queue up screaming “please take my money”. That may be over though, and now the opposite reaction is showing up, treating ML as some kind of poison, regardless of whether it is a simple procedure from the 1990s or an insidious cloud-based service that harvests your data.
In my region, when they add cameras, they even add blue signs saying “The region acts for your security”, which is kinda hilarious (and sad at the same time) because they got bashed years ago already for using public funds to spam villages’ entries with “The region helps its communes”.
Actually, they’re quite correct about this. Denoising, superresolution, segmentation and object recognition, low-light enhancement - these all have pretty clear military-esque applications I should think.
But like, thats true of a lot of things already in darktable that have nothing to do with ML. So unless we’re talking facial recognition I don’t really thing this is much more than a straw-man.
For my part, I am not an artist, but a photographer. More precisely, an amateur photographer.
In my opinion, the use of new technology is beneficial in most cases. However, this is not true in all cases, particularly when personal skills, knowledge, and experience are outsourced to a black box such as AI. A study of medical professionals from various fields shows that those doctors who used AI in their work lost some of their previously acquired knowledge within a very short time. Which I can well imagine, because what you don’t use, you forget. With exceptions, such as riding a bicycle, perhaps.
If it’s just a matter of creating beautiful images ‘somehow’, then in my opinion a smartphone and Snapseed are enough. That’s all you need. But it’s no fun, neither taking the photos nor editing them.
One still must download the code - and if the models grow in size, the darktable binary will grow in size.
Again, that is where we have a fundamental disagreement that cannot be reconciled. I don’t believe in a society where people make individual decisions based on what is best for them because that has led to a variety of disasters and is not sustainable.
Moreover, the complains that artists had about photography may be similar but we must also consider the speed and power. A car crash at 20km/h is similar in concept to one at 200km/h but the latter is what we must watch out for simply because it is more powerful.
Exactly what I disagree with. Of course, people are free to do what they want, but I disagree that we should wait for the technology and then wait to evaluate it. Prudence is important as well, as some technologies can be so disruptive and seductive that we won’t be able to think rationally about them.
Machine learning/AI has already taken away jobs, it allows big tech to concentrate wealth at the top, it is an accelerator for consumerism, etc. It’s dehumanizing.
The algorithms are not the same. However, the ways algorithms influence people are similar and thus we can take a lesson from them.
I hope some brilliant developer with experience in ML and edge detection, creates an efficient add-on to give us all the power to select and mask regions on a photo with our mouse. And I am confident this capability will not cause the world to end.
Since all commercial softwares (Lightroom, Capture One, Luminar, DXO Pholab, ON1 Photo Raw) have already introduced some sort of AI to improve their workflow regarding noise reduction, sharpening, masking etc etc I suppose the world population will survive
Joking aside, as already reported previously on this thread there are also some open source softwares which are taking this same AI route: Kdenlive for masking, Shotcut for text and voice recognition, Digikam for culling pictures, face recognition etc
Reading the very interesting posts on this thread they all reminded me some similar concerns (fear of losing job by artists mostly) I have noticed on the Krita GitLab web-page.
At present, for Krita there is an external plugin for AI.
I am 100% positive that whenever, in the future, the darktable developers might decide to introduce some sort of AI this option will not be “enforced” on their users. In short, they will not be forced to use it with dartkable or to download many GB of data etc etc
Disclaimer, personally, I am basically a RawTherapee user not a darktable one. With RawTherapee the masking stuff is pretty poor IMHO (at least compared to ART). I would be very in favour to have some sort of AI masking for RawTherapee as well. Having hundreds of pictures to “manually correct” on a daily basis is not fun at all and is very time consuming…
The problem with AI masking in Darktable is that Darktable does not support arbitrary pixel masks, and has no way of saving them to the XMPs (and no way of keeping XMP file sizes from exploding if it did).
Therefore, at the moment, AI masking would have to result in a vector mask. A prototype implementation already exists, but AFAIK is not finished.
I think there’s room for a separation of tools – some modules in darktable may not even be available in the distant future or may be implemented differently. I see no reason to avoid offering a feature now just because it may be different later. Non-destructive and repeatable are orthogonal issues.
This is the issue. It’s not a black or white issue. One can use AI selectively. Or not at all.
I can see plenty of useful applications in the photo world. My Synology NAS can tag faces automatically. Would love to have that function for other motives too, cars, sunsets, food…
)
