AI image enhancement, good or bad for imagers

Hi William

This is a very interesting topic and I'm sure it will generate a lot of disparate opinions. I think it is too complicated though for a yes or no, so here are some thoughts to ponder and perhaps provoke.

I’m going to argue that first AB is mostly focused on a reasonable scientific realism, for the most part anyway. By scientific realism I mean that we don’t deliberately distort the content, we try to show some real aspect, though it may be hard to define for a NB image, since we don’t actually see things that way. Still even then there is a mathematical relationship between the transformations we do in a NB image and the final result, so it does relate directly to the real image. In this case we do try to be careful how much manipulation we do with AI, as generally we are trying to use AI to remove noise or bloated stars due to seeing, which are noises that get in the way of interpreting the data. Or sharpening to highlight fine detail, but not create detail that is not there. I’m sure in the future AI could do stretches better than we can, but of course, the imager needs to input what they want to show, and how they want to show it, as there are an infinite number of ways to stretch, depending on what you want to show. And so on for other operations we do manually now. I think use of AI here is fine as long as it keeps to that mantra of basically sticking to the reality in the data, and keeping artistic control of it. There is art here, as the resulting image needs to convey something of interest and something the imager is try to say or show. That is always an important criteria, whether one uses AI or not.

I would argue for Art in the most general way, there is no limit on using AI at all, since it is just generating the look and feeling that the artist wants. One is still in control of it and responsible for it, but the artist can and will use any tool available and do anything he wants with that tool. Now if everyone uses AI and all the images look the same as a result, that is not art—if there is no reflection of the artist in the art. I’m sure in the early days of photography, painters said photography was not art at all. And I think it took a long time for that to turn around, for it to be recognized there was art there in photography for sure—no one doubts that now. So generally artists can use AI anyway they wish, but it is only art if in the end it conveys something interesting/disturbing/thought provoking to the audience.

Again, we don’t see unrestrained art on AB, we see art that is tied reasonably to some scientific reality—so here we use AI as long as it sticks to that reality.

Rick

Astro Jeep · Dec 18, 2025, 05:02 AM

I think that AI is perfectly fine for taking slightly oblong stars and making them round. I also think it is fine for removing satellite trails too, and removing NB stars to be replaced with RGB stars. Anything outside of the basic accepted tasks I think folks should state that they used AI to manipulate the image especially if it is being considered as an IOTD. I don’t think an overly AI modified image should be kept from consideration for an IOTD, but I wouldn’t be thrilled to know an image was awarded an honor, where the author hid the use of AI for color manipulation, etc.

Interesting point and your view seems to look to be where I Sit at the moment… for me I cannot afford top end equipment such as ed optics etc, I use an Evostar 102/F9.98. but that means I have to accept the limitations of the equipment as we all do and try for the best my equipment can achieve, so I accept vignetting oblong stars at the corner etc, AI could remove all of that and enhance the image using the best photographs on the net for reference, but could I claim it as my own?. I do think it should be labelled as AI enhanced and reference made to AI in the software section of the image details. filling this section should be best practice anyway.

William Sweeney · Dec 18, 2025 at 09:39 AM

Astro Jeep · Dec 18, 2025, 05:02 AM

I think that AI is perfectly fine for taking slightly oblong stars and making them round. I also think it is fine for removing satellite trails too, and removing NB stars to be replaced with RGB stars. Anything outside of the basic accepted tasks I think folks should state that they used AI to manipulate the image especially if it is being considered as an IOTD. I don’t think an overly AI modified image should be kept from consideration for an IOTD, but I wouldn’t be thrilled to know an image was awarded an honor, where the author hid the use of AI for color manipulation, etc.

Interesting point and your view seems to look to be where I Sit at the moment… for me I cannot afford top end equipment such as ed optics etc, I use an Evostar 102/F9.98. but that means I have to accept the limitations of the equipment as we all do and try for the best my equipment can achieve, so I accept vignetting oblong stars at the corner etc, AI could remove all of that and enhance the image using the best photographs on the net for reference, but could I claim it as my own?. I do think it should be labelled as AI enhanced and reference made to AI in the software section of the image details. filling this section should be best practice anyway.

I think you could claim an image as your own after fixing oblong stars. I am still fairly new to this, but I think you set the amount of correction to be made and then run the tool. You made the decision to use a tool, just like you made the decision to use your telescope which is a tool. We can run down the rabbit hole here, but I liken it to the Supreme Court’s Justice Stewart’s subjective standard for determining obscenity. He couldn’t define exactly what obscenity was, but he said, "I know it when I see it,” and I think that is how we have to view the use of AI acceptability in just about everything.

I my view, AI is just a tool like many others we use in processing. It can be very powerful and is already useful in giving us better tools to do things like noise reduction, aberration correction and sharpening. Some of our tools, such as star removal could work a lot better so there’s still room for improvement. A lot of the tools we use could be smarter and if AI can provide that, so much the better. The main thing is having control over the process. Imagine an AI tool that just says “click to process” to generate the best possible image from your data and it really did, what then? From what I know of human nature I suspect a lot of people would use it. I can see a world where smart telescopes pick the subject, take the data and send a perfectly processed image to your phone. The huge threads you see anytime smartphone astrophography is discussed on Cloudy Nights tells you how popular a system like this would be.

Just as I would have said to darkroom based photographers in the 70’s, enjoy it while you can.

Bill McLaughlin:
Good results should be earned, value comes from labor, and if something becomes too easy, it becomes trivial and worthless.

For noise reduction, deconvolution, and mask generation AI is great. When it starts creating things that are not real is where I would draw the line.

Ricardo Serpell · Dec 18, 2025, 02:22 PM

No more denoising my images : I feel noise is a natural component of a real photograph (sort of a part of its being)

OK, I feel noise is a problem when we draw with light using the technical apparatus we have and so I agree in using a bit of noise killing tool but how much noise is really introduced by our apparatus and how much noise is in the blackest part of our DSO images coming from unresolved starlight ¿?

🤔

Look at daylight photography with analog film where the noise comes from the film grain. We humans do not see noise apart from the floaters in our eyes which maybe develop more and more due to aging of our eyes…

Rick Veregin · Dec 19, 2025, 01:00 AM

Hi everyone

This is a fascinating discussion about how we can and should use AI.

I do want to make a point I missed making in my initial post, and so far I haven’t see it discussed.

What fundamentally is AI?

In the 80’s as part of my work, we began experimenting with Neural Networks to find optimal solutions for our data. To be clear I did not write code, I just used it. Instead of trying to use linear, or logarithmic, or quadratic, or spline trends to find optimal solutions, the Neural Net used nodes with coefficients, where the coefficients were created by training on a set of data. Then once trained, one could find optimal solutions for that data, and make future predictions, and of course, with new data the Neural Net was refined. This was the early days of what we call AI, but the basic principles remain the same today. AI creates some sort of neural network that fits the training data.

The key point I want to make though, is Neural Networks and thus AI, are just algorithms—mathematical transformations. Much more complicated algorithms than we are used to, and more complex than we can comprehend, but still algorithms.

That means AI is in the same class as a digital development stretch, a hyperbolic stretch, an SHO palette—even the debayer on an OSC camera, or a colour calibration, all are algorithms that we use without thinking or understanding.

What I’m going to argue is that all that we do already depends on algorithms, and AI is just another algorithm—maybe one that we don’t understand, but how many can say they truly understand a hyperbolic stretch?

Fundamentally, then AI is a set of algorithms that explain the data, a set of mathematical equations. We may not understand the math, but that is nothing new for us.

So AI is just a continuation of the algorithms we must use to make an image of any kind. It may seem scarier, but anyone can butcher data with the current algorithms, I’ve done it myself—it doesn’t take AI to mess with data. I do think the term AI is just plain wrong and over-hyped—AI is a complex algorithm that doesn’t think at all. And it makes us scared that is intelligent and will replace us. Okay it may replace us is some tasks, but what we have now is not intelligence. And progress on understanding how our brain works and intelligence is to my view as far away as ever.

Rick

Thanks Rick, there has been some really insightful posts on this topic, and a lot of them like this one really stretch my understanding of the topic, which is great. I do use chatgpt a lot, but not so much with the images themselves, I am still learning about a lot of the different aspects of the hobby so as an example i wanted to improve my guiding with phd2 so I got onto chatgpt it asked me to upload the phd log and within seconds the algorithm had analysed The data and come back with a set of recommendations to improve the mounts guiding, I further questioned what these changes were doing and was able to learn and reduce future dependence. For me it is like having tech support at hand. I do think and please guys correct me if I am wrong, AI could end up corrupting itself, because as I understand, it sources information from data available on the internet, we all know there is a lot of garbage out there, and some chatgpt responses are not correct. So when AI starts to incorporate it’s previous responses into the data it uses, it learns it’s mistakes and makes them again creating another iteration of error. Oh and finally blurring of the lines. I am already unaware of where some of the programs i use to process images are running AI algorithms, I suppose if i have had a bad nights imaging and I ask AI to enhance the bad photo and it comes back looking like it has just been downloaded from the JWST, then possibly it’s gone too far….

Out of curiosity, I tried it on an image and was pleasantly disappointed in the outcome. Also, don’t know if it was specific to the version of AI that I used, but it was limited to 10MB file size. There are some things I just want to do myself. Also, I believe that some PI processes already use AI such as Russell Croman tools.

Bill McLaughlin · Dec 18, 2025, 03:56 PM

Tony Gondola · Dec 18, 2025, 02:28 PM

The main thing is having control over the process. Imagine an AI tool that just says “click to process” to generate the best possible image from your data and it really did, what then? From what I know of human nature I suspect a lot of people would use it.

Very true. AI in general will result in two things:

One is more content that is no longer mostly “data derived” but instead largely “AI derived”. There is a line between AI enhancing details present in the data and AI “re-imagining” the entire content. AI can often cross that line. This can be seen even now by just sliding the BXT slider to 100 on some images….but…. BXT is actually pretty benign as AI goes since it was trained and developed with a content-knowledgeable human in the loop. I shudder to think what will happen if a well informed human is not there to evaluate the results.

Second is the inevitable “dumbing down” of the humans. People are lazy and will let someone or something do the work way too often. It takes character and determination to avoid that, things not everyone has.

For my part, I see zero wrong with removing stars or noise with AI. One just needs to verify that it is not doing more or less than it should but that is pretty simple and obvious with a quick visual check of the results. In the case of noise, it actually does less to the image than older methods so let’s not labor under the illusion that those older noise reductions somehow did nothing bad to the data!

Enhancement like BXT clearly have the potential to create but with the aforementioned human in the loop it can be kept within the bounds of enhancement and well short of creation. I use it but the more I use it the lower I tend to set the parameters and by turning preview on and off it is not hard to see where the right place is to set the parameter.

The real worry is not AI itself but rather overuse and misuse by the nefarious or the lazy and that is entirely a human problem.

As far as what bothers me personally, it is not the enhancement, since anyone that is an astroimager can tell what is real most of the time. It is the idea that processing that now requires years of experience and creativity will become a single button press. As JFK said: “…..and do the other things, not because they are easy, but because they are hard”. Good results should be earned, value comes from labor, and if something becomes too easy, it becomes trivial and worthless.

Hi Bill, as a child of the 60’s I have strived to live by that JFK quote, when embarking on an endeavor, and as a teacher would often use that quote with my learners who were learning difficult concepts. So in astrophotography my pictures are far from where I want to be, but that’s the learning, I want to learn and use the tools myself to achieve the results I want and while not in the upper echelon of imaging, I get an immense sense of satisfaction when I produce an image that I like while knowing I can still improve. The idea of taking an image and just giving it to a one click creative algorithm makes in my eyes the excersise pointless, to quote your comment “Good results should be earned………”. the problem of a proliferation of AI created images is that It can give a sense of false expectation of what can be achieved. I know folk have been saying it’s easy to discern if an image is AI, and that may be true at the moment, but in the future?……

Rainer Ehlert · Dec 18, 2025, 07:17 PM

“How much is too much” ?

IMHO is when they look like childrens Birthday cakes… 😀

Joke aside. It is good for me to have BlurXTerminator in order to cover up a SDE error never solved by iOptron on my CEM 120EC2 mounts. Without BlurXTerminator I could never show my images.

Noise is another point but here we have to see if noise is noise or perhaps unresolved starlight far beyond the cpanility of our telescopes and this has always raised a question for me, ¿how dark is really the darkest place in deep space?

I read somewhere the SQM value of deep space is 27.42 mag per square arcsecond… 🥴 ¿What value could that be in RGB? for sure it is not 0,0,0 …

I remember when a Granddaughter had her 3rd or 4th birthday. Theme birthdays were the new rage.

She had a Smurf birthday. The cake was all smurf’d out and a lot of blue in the iceing. She basically stayed with us as a rule while her Mom and Dad worked.

The next morning she came running out of the bathroom yelling in panic, “Grandma,Grandma,Grandma! My Poop’s BLUE!”

So not just Birthday cakes… 🤣 She’s 27 now, and an Assistant DA.