Processing in AI (split-off thread)

[pickerdd]

Advertisement (gone after registration)

On 8/27/2025 at 8:19 AM, chris_tribble said:

@pickerdd I second the benefit of diopters on the M. My eyesight has fortunately been stable for many years (🤞) but I’ve used +3.5 diopters on my M bodies for years as I prefer not to wear glasses while out and about.  Of course the specs are on when driving, reading, watching tv etc!

i tried the diopter on the M and it helped but did not solve my issues focusing the rangefinder patch.

[jaapv]

https://mastering-lightroom.com/denoise-lightroom-classic/

[Slender]

1 hour ago, pickerdd said:

 

You addressed the only thing i would like to improve in my SL2 - low light photography. your ISO 12,000 looks pretty dang good!  i have never tried denoise in LR, but i will check it out.

It is game changing. I used to have a love-hate relationship to my SL2 because of my unrealistically high expectations when I got it.... but in retrospect its weird from me because I traded my M9 to jump to the SL2 back then... which is a huge lead in terms of sensor performance in all areas.

My stable today is: SL2, S 007 and M10R.

[Jeff S]

Ditto regarding LR de-noise; that and other LR Classic editing tools continue to improve significantly over time.  A steal at $120 per year IMO.

[chris_tribble]

22 hours ago, Slender said:

With Lr Denoise and outputting your final images at 24 Megapixel (which is still filling up a 6K monitor that, lets face it, most people/clients do not even have), one can confidently use the SL2 at ISO 8000-12500. Highlight recovery becomes uncanny good at those settings. so you can actually expose for shadows (cleaner) and recover your blown highlights in post-production. Its the opposite at ISO 50 (for which you should only expose for highlights and then boost your shadows in post)

You've given me food for thought.  Up until now I've limited myself to 6400 with the SL2 (don't forget I can remember when 1600 ISO Fuji color slide film was the best you could get!).  With lens IS and IBIS it's not felt like too big a limitation - but I might set AUTO ISO a tad higher for the next shoot just to see!

[jaapv]

Personally I prefer the Topaz Photo AI plugin, but everybody should judge for themselves 

[chris_tribble]

Advertisement (gone after registration)

7 hours ago, jaapv said:

Personally I prefer the Topaz Photo AI plugin, but everybody should judge for themselves 

 It tried it.  Does it maintain the dng or produce a tiff?

[jaapv]

I use it in Photoshop as plugin at the very beginning of my workflow when opening from ACR so I would say .psd;  this question does not really come up. Even more so in LR, as the DNG conversion takes place during export, LR being non-destructive.  AFAIK Topaz standalone works with TIFF (but I may be mistaken), unlike DXO which provides the best noise reduction and exports as a new separate DNG.

[Slender]

9 hours ago, chris_tribble said:

You've given me food for thought.  Up until now I've limited myself to 6400 with the SL2 (don't forget I can remember when 1600 ISO Fuji color slide film was the best you could get!).  With lens IS and IBIS it's not felt like too big a limitation - but I might set AUTO ISO a tad higher for the next shoot just to see!

I still leave Auto iso at 6400 because as you said, IBIS does great. Its only if i need more speed for action that i will crack it up one more, or indeed AUTO ISO to go up to 12K.

I used to pay for Topaz as an extra, but i found that LR took over as of today in both practicality, speed and sheer good looking result. I have an M2Pro mac mini and it speed through the files very easily. As of today LR doesnt create an additional dng file anymore, which is nicer. It just does a more advanced and careful de-bayerisation of the files which also yields better details, and less exagerated color aberations like fringing or moire.

My little intuitions tells me that Lr will, one day, apply this level of high quality de-bayering by default upon importing raw files in the future, when they judge most computer would be able to do so for all files at once. For now one can do it on a case-by-case. I use it even on lower iso files now.

@chris_tribble also I would do most export of jpegs as an oversampled 24 megapixel files, where any residual just gets reduced in size and this process of oversampling (you can do it also in Ps using Bi-Cubic sampling) makes for a crispier looking image.

Edited September 12 by Slender

[Jeff S]

Always trade offs. DxO, for instance, is slow to support new Leica gear, and has no support for monochrome cameras.  Adobe may not be the absolute best for a given single task, but LR Classic in particular has proven to continually improve and consistently offer new features, and generally gets the job done for me (except IP for print).  My use of Photoshop has diminished immensely.  With the quality of today’s cameras and lenses, it’s increasingly the case that the user is the limiting factor.

[jaapv]

Interestingly, ON1 seems to be catching up fast with these present-day AI tools ( masking as well) 

[Stuart Richardson]

I was inspired by this discussion to give another shot at the denoise option in lightroom. My opinion of it remains the same. While it can impressively remove noise, I can't get past the unpredictable and uncanny ways in which it modifies the images. Back when I got the SL2, I did an ISO run at my bookshelf, just to see the extent and character of the noise. At sufficiently high noise levels, the denoise function swaps in garbled and misshapen text on the book spines and makes the wood grain unnaturally smooth. Sure, the image is now remarkably free of noise, but it is no longer representative of reality. I tried it on a less demanding picture of some pine trees taken at night against the stars at high ISO. Here it made bare wood colored tips on top of the pines that were not there (it put in a yellowish color instead of green), and shifted the colors of the trees. There was also a band of magenta under the bottom 5% or so of the frame in another photo...just a bar of color. 

I guess my thought is "what is is all in service of?". It seems like this tool is generative AI light. It tries to get rid of noise, but when it is not sure, it just makes something up. To me this means it means it is no longer related to the objective world, nor fully "my photograph" in a meaningful way. I would also note that it is dubious that these edits would be acceptable in certain contexts. I have applied to some photographic grants recently, and all of them have either expressly forbid the use of AI in the photographs or required detailed disclosure of the use of AI tools. I fully support those requirements. At times it feels like there is a fine line between more traditional algorithmic editing tools and the newer AI enhanced versions, but I think there is a larger gulf than it appears at first glance. Rather than trying to clean up the data and separate noise from signal, they are basically looking at billions of stolen images to try to remake your photo into something it thinks you want. For me at least, that is not what I am looking for in editing software. If that means I can't use a photograph I essentially missed because I had to shoot it at ISO 50,000, then so be it.

Edited September 13 by Stuart Richardson

[jaapv]

18 minutes ago, Stuart Richardson said:

I was inspired by this discussion to give another shot at the denoise option in lightroom. My opinion of it remains the same. While it can impressively remove noise, I can't get past the unpredictable and uncanny ways in which it modifies the images. Back when I got the SL2, I did an ISO run at my bookshelf, just to see the extent and character of the noise. At sufficiently high noise levels, the denoise function swaps in garbled and misshapen text on the book spines and makes the wood grain unnaturally smooth. Sure, the image is now remarkably free of noise, but it is no longer representative of reality. I tried it on a less demanding picture of some pine trees taken at night against the stars at high ISO. Here it made bare wood colored tips on top of the pines that were not there (it put in a yellowish color instead of green), and shifted the colors of the trees. There was also a band of magenta under the bottom 5% or so of the frame in another photo...just a bar of color. 

I guess my thought is "what is is all in service of?". It seems like this tool is generative AI light. It tries to get rid of noise, but when it is not sure, it just makes something up. To me this means it means it is no longer related to the objective world, nor fully "my photograph" in a meaningful way. I would also note that it is dubious that these edits would be acceptable in certain contexts. I have applied to some photographic grants recently, and all of them have either expressly forbid the use of AI in the photographs or required detailed disclosure of the use of AI tools. I fully support those requirements. At times it feels like there is a fine line between more traditional algorithmic editing tools and the newer AI enhanced versions, but I think there is a larger gulf than it appears at first glance. Rather than trying to clean up the data and separate noise from signal, they are basically looking at billions of stolen images to try to remake your photo into something it thinks you want. For me at least, that is not what I am looking for in editing software. If that means I can't use a photograph I essentially missed because I had to shoot it at ISO 50,000, then so be it.

Could you post an example of the modifications? 

[Stuart Richardson]

I think the problem is not as much the results, as those are for the most part very impressive. Rather it is what they indicate...the reinterpretation of image data and manipulation of it, as opposed to try make the best image out of the actual data. But anyway, I don't ask that you all agree with me, I just find it problematic for me. I also think it looks unnatural... 

I should say that the mountain and bookshelf are at ISO 50,000, as I was going to test an extreme case. You can see the magenta bar at the bottom of both. The trees are at ISO 12500.

Welcome, dear visitor! As registered member you'd see an image here…

Simply register for free here – We are always happy to welcome new members!

Edited September 13 by Stuart Richardson

[chris_tribble]

A really interesting example.  --- and I'm just marking postgraduate students' dissertations where I feel I'm confronting the same "reinterpretation of [image] data and manipulation of it, as opposed to try make the best [image] out of the actual data" - although in my case the data is students' understanding of the application of linguistic knowledge in language education!.  The worst thing now is having students writing about the impact of AI on academic wrirtng where it's clear that their texts are AI manipulated, but there's no way of proving it.  Scholars need the equivalent of the C2PA standard built into word processing software!!

Edited September 13 by chris_tribble

[Jeff S]

9 hours ago, Stuart Richardson said:

I was inspired by this discussion to give another shot at the denoise option in lightroom. My opinion of it remains the same. While it can impressively remove noise, I can't get past the unpredictable and uncanny ways in which it modifies the images. Back when I got the SL2, I did an ISO run at my bookshelf, just to see the extent and character of the noise. At sufficiently high noise levels, the denoise function swaps in garbled and misshapen text on the book spines and makes the wood grain unnaturally smooth. Sure, the image is now remarkably free of noise, but it is no longer representative of reality. I tried it on a less demanding picture of some pine trees taken at night against the stars at high ISO. Here it made bare wood colored tips on top of the pines that were not there (it put in a yellowish color instead of green), and shifted the colors of the trees. There was also a band of magenta under the bottom 5% or so of the frame in another photo...just a bar of color. 

I guess my thought is "what is is all in service of?". It seems like this tool is generative AI light. It tries to get rid of noise, but when it is not sure, it just makes something up. To me this means it means it is no longer related to the objective world, nor fully "my photograph" in a meaningful way. I would also note that it is dubious that these edits would be acceptable in certain contexts. I have applied to some photographic grants recently, and all of them have either expressly forbid the use of AI in the photographs or required detailed disclosure of the use of AI tools. I fully support those requirements. At times it feels like there is a fine line between more traditional algorithmic editing tools and the newer AI enhanced versions, but I think there is a larger gulf than it appears at first glance. Rather than trying to clean up the data and separate noise from signal, they are basically looking at billions of stolen images to try to remake your photo into something it thinks you want. For me at least, that is not what I am looking for in editing software. If that means I can't use a photograph I essentially missed because I had to shoot it at ISO 50,000, then so be it.

Did you apply sliders and/or other available modifications to better optimize results?

 

[Stuart Richardson]

3 hours ago, Jeff S said:

Did you apply sliders and/or other available modifications to better optimize results?

 

Yes. But the slider is just a blending rate. It does not change the underlying edit, just more or less the opacity. As I noted, the objection is as much an ethical one as a results based one. It can look very good, but fundamentally it is taking your image, estimating what it would look like without noise, and then generating that image, at least in part. You can really get a feel for that if you crank it up to 100. Then it looks truly like an AI generated image, which it is. Lightroom is not trying to hide that. As soon as you touch that setting, the "modified by AI" adjustments symbol turns on. Your image is no longer considered traditionally edited even by Adobe themselves. Again, everyone has their comfort level with AI. Mine is hovering around 0. I just thought it was worth mentioning, particularly as a few of the grants I have recently applied for have specifically called out AI editing and either require disclosure or ban its use altogether. This is exacerbated by AI being a corporate magic word that they all utter in hopes that billions of dollars will appear out of thin air.

This makes it hard to tell what actually is going on: is it akin to the way chatGPT generates images based on billions of stolen images, or is it just a typical algorithm, with some in-house results training (kind of like Nikon did for matrix metering decades ago). So the question is: how was it trained, on whose images was it trained (stolen or not), how does it work, and to what extent is it preserving the actual recorded signal, and to what extent is it creating a convincing facsimile. I know a lot of this is pretty niche philosophical reasoning, but it is important to me as an artist, and also as someone who is trying to participate in a meaningful way in the contemporary art discourse. Objective reality is already under such threat from every angle, I at least want the photos I take myself to be faithful recording of the light that hit the sensor, and not the guesses of a learning machine.

Edited September 14 by Stuart Richardson

[Stuart Richardson]

7 hours ago, chris_tribble said:

A really interesting example.  --- and I'm just marking postgraduate students' dissertations where I feel I'm confronting the same "reinterpretation of [image] data and manipulation of it, as opposed to try make the best [image] out of the actual data" - although in my case the data is students' understanding of the application of linguistic knowledge in language education!.  The worst thing now is having students writing about the impact of AI on academic wrirtng where it's clear that their texts are AI manipulated, but there's no way of proving it.  Scholars need the equivalent of the C2PA standard built into word processing software!!

The last time I was grading students' papers was twenty years ago, and I was barely older than them. At the time, it was typically very easy to tell because the plagiarized material was invariably from something they googled or was on wikipedia, so even if they modified it a bit you could copy/paste it and find it right away. I would tell them this before they did their papers, but it did not make a difference...several would still do it. With AI, I can only imagine how much worse it is. By freeing ourselves from thought, we are only robbing ourselves of a chance to grow as people and stunting our critical reasoning faculties at a time when they have never before been more needed. I can't help but feel like these AI magic wand editing tools are doing the same: freeing us from skill, technique and vision. But anyway, time for me to get off my high horse for the day. I was lucky to get a beautiful autumn walk in the woods with the SL2S. I don't think I will need noise reduction, lol.

[pgh]

1 hour ago, Stuart Richardson said:

This makes it hard to tell what actually is going on: is it akin to the way chatGPT generates images based on billions of stolen images, or is it just a typical algorithm, with some in-house results training (kind of like Nikon did for matrix metering decades ago). So the question is: how was it trained, on whose images was it trained (stolen or not), how does it work, and to what extent is it preserving the actual recorded signal, and to what extent is it creating a convincing facsimile. I know a lot of this is pretty niche philosophical reasoning, but it is important to me as an artist, and also as someone who is trying to participate in a meaningful way in the contemporary art discourse. Objective reality is already under such threat from every angle, I at least want the photos I take myself to be faithful recording of the light that hit the sensor, and not the guesses of a learning machine.

Really getting aside here but this is why I regard the computational photography from most smartphones to be just about the equivalent of horseshit. Nice to look at but about as related to photography as a comic book for me.  

It’s tough, trying to draw these lines, or reckon with them - especially when the world of photography is still so full of those post modernist disciples that really do think that objective or universal truth don’t matter. A true ideological gift to the fascists of the world if there ever was one (and from the strident left, who’d have thought that!?)

But anyways, I’m with you in mind and in practice.

I never use de noising for any of my cameras at all - including my sometimes frustratingly noisy in some ways SL2. Oh well.  Sometimes I forget that it is automatically applied and forget to turn it off. I’m aware there is no perfect capture of reality, but for me I am mostly interested in presenting a fidelity to the record that was made, flawed as it could be. In turn, this represents a sort of fidelity to the subjective nature of the capture itself. If a distorted lens made the record, I’m even inclined to keep the distortion. Everyone has different ways to handle this but what’s important is that it is well and truly considered I think. 

Edited September 14 by pgh

[Advertisement]

Hallo,

Look here: Interesting Leica gear

[jaapv]

TBH I have little interest in in the philosophical aspects of the process and  tend to judge the result