A while ago I found a piece of FOSS software named jpeg-quantsmooth. It essentially attempts to repair the quantization error of a JPEG’s DCT coefficients. What it means for us mere mortals is that it tries to remove JPEG artifacts from an image.
I’ve read the MB cover art doc page, and while this technique is not AI-based, it kinda sounds like cover art “repaired” using it classifies as Fan Art? What do you think?
I’ve used this program on the cover art of two releases.
Flicker (see comments)
Tabou nº2, retour en force (just front and back)
I think the edits you made are very minor and quite within the realm of fixing image quality issues, like removing specks or other blemishes. I also absolutely hate JPEG artifacts (and by extension, JPEG itself), so any edits to remove those blocky edges are a big plus to me.
I think what the wiki page suggests to be inadmissible are things like removing or moving elements that belong on the cover (like the release title). Also, when you upscale an image with AI, you run the risk of the AI making things up that simply aren’t there. And that’s not what we want either.
AI is not some protected certification and because the hypetrain is going so fast, everything is labeled AI at the moment. So I would never say something like: “AI shouldn’t be used, as long as it’s not AI it’s OK”. It’s all about the result of whatever way you are doing things.
4 Likes
One thing is sure, digital releases are a no. While its more of a mathematical process, it is still a modification of the coverart.
I could see it used for scans, but I would rather scan it in a lossless format in the first place
5 Likes
JPGs can be saved with low compression. I use 85% and the difference from a PNG is hard to see on a fancy 4K 24" screen in front of my nose.
If someone is saving 40%JPGs then using an algorithm to bodge it is not ideal. You don’t know what else you loose in the process as you don’t have the original to hand to compare. My personal opinion, it is better to leave the low quality for someone to replace with a better real image.
2 Likes
Yeah, this mostly concerned Discogs releases where no other cover art source is easily available.
1 Like
How about TIFF?
New expenses for server disks. 
The thing with Flicker is that the actual release appears to be totally unavailable today. I have my beliefs that what we currently have for that release are resized versions, and a better version must’ve been available at some point.
I would love to have that, an actual better official version of that cover art, but as far as I’ve looked there isn’t one, so I did this thing.
Next time I’ll keep it for myself 
JPEGs can already look good, I know!!! I upload my scans as JPEGs, and I think they look splendid.
You don’t know what else you loose in the process
While the algorithm obviously doesn’t know right from wrong, it doesn’t simply paint over smudges and dots. It’s grounded in how the codec does compression. It tries to reverse that a bit, and it doesn’t do much more than that. It’s not ideal but much less destructive than smudgy uncanny “AI” solutions.
Anyway, I’ve pretty much moved on from using this technique in the context of MB. IIRC I haven’t used it at all since the examples in this post, and I haven’t felt the need to.
I’m gonna stop necro-ing this, sorry. I didn’t get any notification emails from the forum ^^’