and please use their services
AI in Post Processing
- Thread starter miaamcc
- Start date
- Messages
- 1,918
- Name
- Jamie
- Edit My Images
- Yes
I haven't used it much, but from what I've tested I think it's great for denoising. I'm starting to use it more for masking/object selection - It can still be hit and miss but I'll try it first and if it can't do it right I do it myself.
Content aware fill is just witchcraft now. It blows my mind. Ethically - I'm fine with it in situations where I'd normally use a heal/clone tool anyway - bits of trash in the scene, pimples on faces etc. My "line in the sand" has always been if it's something temporary like trash or a pimple, I'll remove it, but if it's something permanent like a lightpost or a scar, I'd leave it.
Like most new tools, it's easy to overuse it. Just because a slider is there doesn't mean it needs to be used. There is a certain look to photos that have had the AI denoise, sharpen, and upscale treatment, and i'm starting to see it more and more in birding/wildlife circles specifically. It's subtle, but the more you use it, the more you learn to see it, much like the AI "writing style".
Ultimately, it's just another tool. People asked the same question of selective colour or HDR, or even just digital in general - People will overdo it, then it will calm down and become just another tool to subtly shift your images towards your vision.
Content aware fill is just witchcraft now. It blows my mind. Ethically - I'm fine with it in situations where I'd normally use a heal/clone tool anyway - bits of trash in the scene, pimples on faces etc. My "line in the sand" has always been if it's something temporary like trash or a pimple, I'll remove it, but if it's something permanent like a lightpost or a scar, I'd leave it.
Like most new tools, it's easy to overuse it. Just because a slider is there doesn't mean it needs to be used. There is a certain look to photos that have had the AI denoise, sharpen, and upscale treatment, and i'm starting to see it more and more in birding/wildlife circles specifically. It's subtle, but the more you use it, the more you learn to see it, much like the AI "writing style".
Ultimately, it's just another tool. People asked the same question of selective colour or HDR, or even just digital in general - People will overdo it, then it will calm down and become just another tool to subtly shift your images towards your vision.
- Messages
- 4,726
- Name
- Graham
- Edit My Images
- No
There are several other threads on this topic. Some are specific to particular applications, but others are more general.
I just searched on "AI" across all forums, which seemed to bring up a good range of options worth following up
hashcake
Gone to pot!
- Messages
- 5,948
- Name
- Darran, Daz or ****
- Edit My Images
- Yes
I recently had a play with AI after I'd processed photos.
I was quite impressed but you could tell AI had been involved.
I'll be honest and admit that I find PP a laborious task
So I'm lazy and will initially use auto in LR.
After that I'll make a few tweaks if I feel they need them.
I couldn't find an AI that would accept raw files as that was what I initially was intreaged about.
I was quite impressed but you could tell AI had been involved.
I'll be honest and admit that I find PP a laborious task
So I'm lazy and will initially use auto in LR.
After that I'll make a few tweaks if I feel they need them.
I couldn't find an AI that would accept raw files as that was what I initially was intreaged about.
- Messages
- 863
- Name
- Phil
- Edit My Images
- No
As part of my import preset in Lightroom I have AI dust removal which is excellent and a great time saver.
- Messages
- 1,126
- Name
- John
- Edit My Images
- No
The problem is 'AI' is an extremely general term that covers a very wide range of technologies and many features and processes you've been using for years are already technically 'AI'. When I studied AI in the late 90s and early 2000s, some of the techniques went back decades since at the time one of the main focuses was improving performance with limited hardware especially in areas where conventional computing had no faster answers like constraint satisfaction problems (like building a timetable with minimal conflicts)
These days the term AI is often used specifically for Generative AI (genAI) using Large Language Models (LLMs) which ironically are in many ways the opposite of traditional AI because they're extremely inefficient and wasteful but actually very stupid, they're using probability models to look smarter than they are. There are also significant ethical problems with genAI since they rely on being able to steal huge amounts of other peoples work for them to function which isn't an aspect of many other AI systems.
In terms of post processing, say for example you want to improve the detail on a leaf in an image. You can use traditional AI upscaling techniques that try to work out what colours the extra pixels should be which likely won't work if the image is particularly low resolution or you can can use a genAI tool which could generate an image of a leaf which matches the dimensions in the image using data it's stolen of other leaves. The latter will look much better but it's clearly not your image any more and that to me is a stage too far.
These days the term AI is often used specifically for Generative AI (genAI) using Large Language Models (LLMs) which ironically are in many ways the opposite of traditional AI because they're extremely inefficient and wasteful but actually very stupid, they're using probability models to look smarter than they are. There are also significant ethical problems with genAI since they rely on being able to steal huge amounts of other peoples work for them to function which isn't an aspect of many other AI systems.
In terms of post processing, say for example you want to improve the detail on a leaf in an image. You can use traditional AI upscaling techniques that try to work out what colours the extra pixels should be which likely won't work if the image is particularly low resolution or you can can use a genAI tool which could generate an image of a leaf which matches the dimensions in the image using data it's stolen of other leaves. The latter will look much better but it's clearly not your image any more and that to me is a stage too far.
Last edited:
The denoise is a godsend for me since I'm often shooting sports indoors at 6k/8k/10k iso. Less often but a great time saver I'll use the AI masking. Content aware fill is a path of last resort for me, I try to set things up so I don't need to edit content in an image, so it get used if I mess up.
- Messages
- 515
- Name
- David
- Edit My Images
- Yes
Have a look at ON1 Photoraw and try some of the AI masking and one click improvements. Even if its not quite what you want (do American viewers have different requirements?) it provides a good start to post processing. So much easier than tracing round the outline of a subject as we did in the old days.
- Messages
- 3,972
- Name
- Wayne
- Edit My Images
- No
Shoot Film
- Messages
- 12,272
- Edit My Images
- No
Frankly, I blame John McCarthy, Steve Russell and the rest of the LISP gang.
In the 1980s I was on a project that did exactly that. It wasn't my baby but the guys who did it used "C" and got it running on time and to specification. I would be interested to see if generative AI would be quicker, cheaper or better.
I agree. My wife just showed me an image that is being presented as a photograph but is quite certainly a generated image. Material like that will certainly be used to promulgate more and more lies in a world already awash with untruth.
Last edited:
Topsy
Challenge Owner
- Messages
- 4,243
- Name
- Paul
- Edit My Images
- No
It takes the work out of dust removal (I use LrC) and is often quicker than the old Clone Stamp Tool for temporary distractions like litter etc or even people in the landscape who are transiting but not quick enough for me to wait them out but that's where it ends for me. I wouldn't try to use it to add something that wasn't there at the time os as has been said remove things that are permanently there.
- Messages
- 1,736
- Name
- Peter
- Edit My Images
- Yes
Ah, LISP, my favourite programming language of all time!Frankly, I blame John McCarthy, Steve Russell and the rest of the LISP gang.
In the 1980s I was on a project that did exactly that. It wasn't my baby but the guys who did it used "C" and got it running on time and to specification. I would be interested to see if generative AI would be quicker, cheaper or better.
I agree. My wife just showed me an image that is being presented as a photograph but is quite certainly a generated image. Material like that will certainly be used to promulgate more and more lies in a world already awash with untruth.
I had Donald Michie's book which described various AI techniques in the context of the LISP language, and the AI edition of Byte magazine, 1982 I think it was (subsequently gifted to the computer museum in Leigh Spinners Mill as was my BBC Micro LISP interpreter manual).
My main use of the language was in customising AutoCAD with AutoCAD's AutoLISP interface.
I missed out on LISP. We had the BBC Micro at home in the 80s but went the C/C++ route, and my brother ultimately went on to study neural networks for his PhD, applied to fingerprint recognition.
- Messages
- 3,740
- Name
- Tim
- Edit My Images
- Yes
Gen AI is terrible at complex timetabling; that's not its strength. You need different "AI" for that, called "Operations Research" in my day but rebadged as AI today.
I find that's my main use .... feels very mundane. But I think it's useful. Usually doing something like taking a 3:2 and extending it to a 5:4 or 4:3 for actual final use.
I think AI can also sometimes useful to help with masking,
I think cheaper ... possibly. 'better' is subjective.
Some of the criticisms of what AI generates in the way of code will in part being down to modern coding and deployment practices. The last 20 to 25 years have seen more and more bloated and inefficient and slower software. If the AI is trained on that it will naturally produce more of the same.