Hi all,
please take this post with a grain of salt. I am not affiliated with any developers of AI programs whatsoever. I am also not a fan of AI tools in astro imagery. I have seen quite some images lately, that show too heavy or wrong use of deconvolution leading to "worm-like" features of small details in images.
However, with the recent uprise of AI helper tools within this community of astrophotographers I think we can now all take the last step and let AI generate the image
. I mean we do have remote controlled scopes all over the world. Download the sub exposure and have to fiddle with post processing... 

I learnt of midjourney v5 by chance today and was intrigued and fiddled around with it. After generating various images of whatever I wondered how good it is at generating astroimages. So the prompt I used was: A photograph of the andromeda galaxy through a telescope with 150mm aperture and 1000 mm focal length. The photo must have at least 20 hours exposure time. The neighbouring galaxies M32 and M110 must be within the field of view.
Midjourney would generated 4 images. Again these are not photos collected in a google search on the internet, but artifically generated images:

Now clearly, this is not the correct representation of M31 to begin with. If you are familiar with M31 you will notice that the top right image has the correct features: reddish core, blue outer arms with clusters of stars and some H-aplha regions.
I continued to use #4 (bottom right) to shuffle and regenerate images:

I really like photo in bottom left. Nice stars really. Maybe a little too much sharpening without properly applied mask for the stars ;).

And finally another set of images:

Looking quite nice doesn't it? Still not M31 but estatically pleasing. Collimation might be a bit off when looking at the stars, so the AI needs to learn to collimate the scopes properly first
.
I am amazed, also floored by what is possible with AI image generation. The above image is of course not correct to the description, which is probably due to niche data or lacking description. However, there are virtually no limits what you can do currently with it. Also questions about copyright and ownership will be interesting.
With respect to the recent advant of AI tools for image (re-)construction in astro-pictures I am wondering if the AI can be used to be fed with the 100s of exposures and do a fully automated image processing. Oh wait, we have that...
.
Also if you still want proper data to work with just talk to Alexa and instruct her to let ChatGPT controll your Unistellar. Go to bed and next morning you have your image. How much fun is that
.
Br, and keep imaging please ;)
Michael
please take this post with a grain of salt. I am not affiliated with any developers of AI programs whatsoever. I am also not a fan of AI tools in astro imagery. I have seen quite some images lately, that show too heavy or wrong use of deconvolution leading to "worm-like" features of small details in images.
However, with the recent uprise of AI helper tools within this community of astrophotographers I think we can now all take the last step and let AI generate the image



I learnt of midjourney v5 by chance today and was intrigued and fiddled around with it. After generating various images of whatever I wondered how good it is at generating astroimages. So the prompt I used was: A photograph of the andromeda galaxy through a telescope with 150mm aperture and 1000 mm focal length. The photo must have at least 20 hours exposure time. The neighbouring galaxies M32 and M110 must be within the field of view.
Midjourney would generated 4 images. Again these are not photos collected in a google search on the internet, but artifically generated images:

Now clearly, this is not the correct representation of M31 to begin with. If you are familiar with M31 you will notice that the top right image has the correct features: reddish core, blue outer arms with clusters of stars and some H-aplha regions.
I continued to use #4 (bottom right) to shuffle and regenerate images:

I really like photo in bottom left. Nice stars really. Maybe a little too much sharpening without properly applied mask for the stars ;).

And finally another set of images:

Looking quite nice doesn't it? Still not M31 but estatically pleasing. Collimation might be a bit off when looking at the stars, so the AI needs to learn to collimate the scopes properly first

I am amazed, also floored by what is possible with AI image generation. The above image is of course not correct to the description, which is probably due to niche data or lacking description. However, there are virtually no limits what you can do currently with it. Also questions about copyright and ownership will be interesting.
With respect to the recent advant of AI tools for image (re-)construction in astro-pictures I am wondering if the AI can be used to be fed with the 100s of exposures and do a fully automated image processing. Oh wait, we have that...

Also if you still want proper data to work with just talk to Alexa and instruct her to let ChatGPT controll your Unistellar. Go to bed and next morning you have your image. How much fun is that

Br, and keep imaging please ;)
Michael