Updated review, working much better after resolving some user error. I would edit my previous post if I was allowed to do so.
I solved my file access problems by fixing my OneDrive backups, which were apparently creating parallel problems unrelated to Cosmic Clarity. Files seem to work just fine now.
Regarding the reports that it's slow, it seems fine to me. Denoising is always kind of slow, but the non-PI interface gives you a window with a percentage countdown so it's very tolerable. Takes maybe 30 seconds per image on my system.
The standalone interface isn't great, because you have to manually move your file into the input folder, run CosmicClarity, then navigate to the output folder, copy the result and paste it back into the input folder and run CosmicClarity Denoise. This does allow you to play with the settings for each step, but once you have those dialed in it would be nice to run them both sequentially from the same window.
And the results are definitely promising. As an example, this was my entry into the monthly imaging challenge, using my old workflow:

And here are some zoomed in crops of the core of M33, the original on the left (from the above) and on the right is a version I reprocessed using CC and CCDenoise (between background extraction and stretching).


You'll have to excuse difference in color balance, brightness, etc because this was an entirely new reprocess of the same data. I just manually cropped the centers of each image so they're not a pixel perfect match but they're pretty close. I think the improvements in star size and shape are pretty obvious, the dust lanes are at least nominally improved, and I think the denoising is better than what I previously got out of Graxpert's 3.0.2 AI denoise. The original image was about a 50% crop from a C8 hyperstar system with an ASI2600mc.
If I really stare at these two images, I can see some of the same problems that BXT has, in terms of a seemingly random determination of what is a tiny star to delete as background noise and what is noise to enhance into a tiny star, but it seems much less aggressive about making things up than BXT is. I will continue to play with the settings to see if I can't improve my results.
But even this first pass, with just the default settings, makes it pretty clear that CC and CCdenoise are definitely superior to my old workflow of deconvolution in Siril and denoising in Graxpert. And it's hard to argue with free. Thanks, dev!
Edited by kg7, 02 November 2024 - 12:39 AM.