Best-looking stars? [Deep Sky] Processing techniques · Emmanuel Valin · ... · 68 · 3516 · 7

MaksPower:
The moment you start using BlurX and the rest it’s all artificial - you may as well pull out the crayons and watercolours.

Pretty picture perhaps but little to do with reality.

You could pick any DSO you like and paste a bunch of randoms dots over it and very few would have a clue whether that was the real star field, or not. With AI few will even know what’s real vs what’s not in a few years time. You won’t even need a rig - just type in “make me an image of <favourite object>” and it will fake it, complete with pinpoint (artificial) stars.

Let's expand on your take. How about the minute you use anything other than film, it becomes fake. You stretch it... fake. You denoise it... fake. You filter it... fake. You remove the gradients... fake. There are so few purists left in the world. I commend you!

andrea tasselli:
Dodge and burn were used even in the good ol' times. And few more tricks besides... I can assure you film was no bar.

+1
I would add hydrogen soaking and negative masking to that list.

This is the quintessential nature of AI in photography of any kind… Not just astrophotography. It really starts to blur the lines between a "real" photo and a "fabricated" photo. 

Now, I've been a photographer for a long time, mid 2000s. Sharpening, noise reduction, maybe some masking and tuning to bring down a nearly-blown out sky… I wouldn't say these fundamentally change the nature of the scene being photographed, but they are editing the QUALITY FACTORS of said photograph. There are, however, many AI "photos" out there that are 100% AI generated, without even being some original source photo just augmented by AI (and those exist too…I've recently been going through lots of landscape photos, where there are notable foreground elements, such as leaves on the ground, which was a very popular fall theme…that are totally fabricated). These 100% AI generated images are often very good, as unlike with humans or animals where you usually end up with an extra digit or few (or too few!), landscapes…often…come out looking very real. So now, every day, when I go through my streams of nature photography….I am asking myself with just about every photo: IS THIS REAL??? My absolute love of nature photography, has been crumbling into a daily dose of uncertainty, confusion, doubt, and often distaste.

With AI, which we are all now being exposed to on a nearly continual basis these days, it is becoming harder and harder to identify what is real and what is not… This was always a personal concern of mine with regards to AI. I've definitely seen astrophotos generated entirely with AI…most are pretty pitiful, because AI is currently rather pitiful at reproducing images of space, or for that matter of Aurora (THANKFULLY!!) A lot of these fake astrophotos are of "nebula" that simply do not exist, and when you are an astrophotographer yourself you generally know about most DSOs, and when you don't, you can then usually go look up if a nebula is real or not, even if its rather obscure. 

At some point, though…I think we are indeed going to reach a point where AI use in photography, including astrophotography, is going to get good enough that it will be really hard to tell if what we are looking at is real or not. When are we going to switch from removing stars, editing the nebula, and adding the original stars back….with generating 100% fabricated starfields from scratch? Are we already there (I was looking at some photos the other day that said they were AI generated RGB stars in narrow band images….!!!!) At what point to we start re-generating, or even generating from scratch, nebula or galaxy structure? I don't think I've seen anything like that myself yet…but, when do we get to that point? When does AI actually get GOOD at generating space images from scratch, such that even skilled astrophotographers have to think hard and maybe be left unsure, of whether a "photo" is real or not?

I think the day is coming… Its starting with stars. BXT is really nice right now…it does the job of deconvolution, but does it so much better, with far, FAR, FAAAR less headache!! Boy, I spent hours in the past, fiddling with deconvolution and regularization settings, trying, tweaking, trying again, to attempt to dial in just the right settings to correct my stars without leaving artifacts strewn about my images. I try to use BXT with moderation, so that I correct and maybe shrink a little, without overdoing it… But its so easy to take it overboard!! Same with other AI powered processing tools. There are undeniable benefits, for sure, and they do make processing astro images a bit easier, and less wholly time consuming. 

I personally, am having a really tough time with AI in a lot of other photography that interests me. Today, you can often, even if it might require some closer scrutinization, tell when an image is AI generated. AI isn't perfect…yet. I do fear the day is not long off, though, especially once we marry quantum computing and LLM tech, where AI will be able to generate images you just can't tell are fake. I like to know that what I'm looking at is real, and exists somewhere on earth, such that I could one day go visit that place myself! The last few months to a year, though, have left me with an ever growing sense of "Is this real or not??" and I now no longer know, a lot of the time, if what I am looking at exists physically on earth, or is some set of remixed and regurgitated pixels out of an LLM somewhere. The lines of reality seem to be blurring, and that's, at least for me, rather unsettling. 

I do fear the day when astrophotography shifts from primarily a complex endeavor of skilled and dedicated photographers spending real money, and real time, to find and expose REAL things in space, to the benefit of everyone who then observes those images….to an industry of wholly fabricated….lies…..

I don't think it is unreasonable to have concerns about AI usage in the fields of photography, astrophotography included. There are very real and legitimate concerns to be had about AI's infiltration into these realms. I think any realm of creative human endeavor….I see them very rapidly being replaced by prompt, point, click………

AstRobert:

Jon Rista:

AstRobert:
I really love these stars here:
https://www.astrobin.com/ahxx7q/?q=Charles%20Hagen

I am actually looking for tips on how to process stars to have nice colors. All of these pictures with amazing stars have extremely long exposure times and i wonder how they manage to not totally destroy the star colors with that. Are you aware of good tutorials for amazing star colors!?

Cheers, Robert

They aren't necessarily all that long, considering apertures and pixel sizes. Under dark skies, with small pixels and narrow band filters it is not unusual to need and benefit from longer exposures. At 300 seconds, just 5 minutes, that is not really all that long. If you had huge pixels, then yes, you might saturate, but with small pixels the stellar signals should be getting distributed over many pixels. 

I have used 10 minute exposures with an f/4 camera lens with a 150mm aperture using narrow band filters, and 2.4 micron pixels. I would usually clip half a dozen to a dozen of the brightest stars, just a smidge. Based on my analyses at the time, the next brightest stars would have allowed another 180-300 seconds of exposure before clipping.

Well I was looking at the RGB exposure times without NB Filters (cause thats a whole other topic regarding star colors ;-) and I get saturated star cores at 120sec with an F7.5 Scope so that is why I wondered, how these great stars colors with 600s at f4 or whatever come to be.

Take separate images at a shorter exposure and the process the stars separately.  I use 30s for stars.  I take about 30 frames per R, G, and B, and then combine them at the end of processing.

Emmanuel Valin:
In my quest to keep improving my images, I’d like to ask the AB community for examples of deep-sky images you find have some of the best-looking stars.

Next step will be about understanding the gear / acquisition & processing techniques that can produce such results, but my first objective is to actually find some images that display top-notch stars according to you.

Of course highly subjective, but that’s what makes it fun !

CS

Emmanuel

One of the useful things about the IOTD awards is that they often produce some pretty good images that serve as examples for aspiring imagers.  Those images  are useful examples for figuring out how to produce a "good image".  Have you looked through the IOTD archives to see what the stars look like?  Very few IOTD awards go to images that have lousy star images.

John

Jon Rista:

Kyle Goodwin:

Oh and one last thing... the images you see posted here are all processed and corrected images. Most of them make extensive use of tools like Blur Xterminator to correct the stars. In my oppinion that's borderline cheating, but to each their own. We all like the pretty pictures at the end.

I don't see anything "cheating" about BlurX, however I personally don't generally use it on my stars only images.  Here's the reason: it's tricky to get BlurX to improve stars without increasing the number of stars which are clipped.  Clipped stars lose their color in the core.  I pride myself on producing excellent stars out of the camera so that I don't need to use BlurX on them (I certainly use it on the starless structure) and I can keep all the wonderful color I've captured.  If you have stars which are less than perfect, but you want to keep as much color as possible you can use BlurX in correct only mode to fix the shapes without shrinking them.  If you must shrink the stars with BlurX because they're bloated due to whatever reason then you may play with the star halo adjustment to try to retain color as much as possible.  Definitely a case of nothing beats good data, though.

If you are using BlurX because you want to "de-emphasize" stars, I'd suggest just stretching the stars only image less before blending it back into the starless image instead.

I think using BlurX is a misrepresentation of the data you captured. Therefore I call it borderline cheating. It's like the whole controversy of the William Optics MiniCat51 all over again.

I don't know that I would go this far. BlurX is not that dissimilar from deconvolution. Is deconvolution misrepresentation, too? Stars are convolved by the atmosphere and optics. That convolution means they are not accurate representations either. Now, I don't generally like how far BlurX is often taken, reducing most if not all stars to roughly single-pixel points.  I think that loses some representation of realistic factors...stars, either due to distance or physical size, do have some diversity in the real world, and I think it is good to try and preserve a certain level of those aspects. 

BlurX differs from deconvolution, in that it is doing the same job, but using AI instead of a strict, iterative algorithmic process. The algorithmic processes of deconvolution are often error prone, and those errors tend to produce artifacts that are also not realistic and in a sense, would misrepresent the true nature of the field you have imaged. Used with moderation, I think BlurX does a better job of achieving the same goal that deconvolution aims to attain: To roll back or reverse the convolution that occurred during imaging, restoring a degree of the "original" nature of the stars. BlurX relies on a different mechanism to achieve that goal. Any tool can be abused, and I think its a lot easier to abuse BlurX for sure...that does not negate the potential benefits of producing a more faithfully representative image that BlurX can offer, however, if it is used effectively and in moderation.

That's why I shoot seperate subs for the stars, with much shorter exposures and only select the 25-50% best of the subframes selected for roundness and FWHM to be stacked. This is somewhat like lucky imaging and makes deconvolution or BlurX unnecessary and is a much better representation of what you would expect stars to look like.

Yeah, shorter subs should give better baseline stars. 

I haven't really done this much myself, as I have never liked the destruction of details around stars that are removed, when you do star removal. Nebula structure and detail is of paramount importance to me, and I've done both manual star removal in the past (using various PixInsight tools), as well as SXT, and there just seems to be too much detail destruction. One of the reasons I do astrophotography is somewhat as a personal scientific discovery. I like finding all the nuances in the details of the structure I image, or finding some obscure super-distant background galaxy, stuff like that. Star removal, at least any time I've done it, seems to hamper that discovery of real world details. 

Maybe some day, there will be an AI star removal tool that can remove stars without any of this detail destruction. Until then, I've used BXT. Its not perfect, but, it does a heck of a lot better job, than my manual deconvolution with Richardson-Lucy. I try to maintain a moderated hand with it, and reduce only to the point that stars don't overpower the extended object details.

One thing you can always do, if you feel its important, is expand the DR of your images first, before using BXT. This can help limit or eliminate unwanted clipping of stars as part of the process. Like with any deconvolution, the process generally aims to gather signal power that was thrown away from the centroid, back into the centroid, so with either BXT or more standard deconvolution routines, this kind of necessitates increasing the intensity of deconvolved or otherwise recovered stars. The energy is being un-convolved, or reconcentrated, back into the centroid. Can't help but make stars brighter when you do that. ;)