I recently flagged post #1148023 and post #1148022 for being inferior to post #3897885 and post #3897810 respectively (they are almost 2/3 times bigger), but the much smaller Tumblr samples ended up deleted instead because the Twitter versions had some compression artifacts on account of them being from Twitter. I was told by the Janitor who deleted them that "[e621 does] not go by resolution" when choosing which version to keep, but this is directly at odds with the flag page which lists being larger as a reason to keep one version over another.

When I asked, the same Janitor also disclosed that they might keep even a 500x500 PNG from Pixiv over a 3000x3000 Twitter JPG, which is just absurd. Being over 60% bigger should already be enough to offset such compression, but apparently even if those images were 500% bigger they still wouldn't have made the cut. I could understand if the image was only, say, 10% bigger, or if the compression was akin to some deep-fried meme, but completely ordinary run-of-the-mill Twitter compression isn't nearly enough justification to ignore such a substantial size increase.

The flag for deletion wiki page and the "Duplicate or inferior version of another post" option on the New Flag page need to go into more detail in regards to what actually constitutes as an inferior version, the staff needs to be consistent with said rules when selecting which post to keep, and substantially larger images should not be thrown out for mere Twitter compression.

Compression artifacts play into when a post is considered inferior or not. Resolution can play a part, but not exclusively; a smaller PNG may be kept over a larger JPG since a PNG is lossless and provides a cleaner image and JPG is lossy and can have minor to severe compression artifacts (of course, it's also possible for PNG to exhibit compression artifacts if it was originally a JPG that got converted to PNG, in which case the original JPG would be preferred as it would have no additional artifacting). Twitter in particular also has a habit of resizing images to have a 2048 or 4096 width or height, which causes it to further deviate from the original on top of recompressing it as a lossy JPG.

I'm not smart enough to argue my thoughts properly, so here's a visual example I hope explains things.

Here's an original (sfw) drawing of mine, featuring 3 versions.
The first one being the original file.
The second one being a slightly reduced version to accomodate FA's old 1280 resolution limitation.
The third one being the original file, uploaded and redownloaded from twitter right now.
https://imgur.com/a/GIuyuxL

If you open all three images on individual tabs, and switch over them, you will noticed that the twitter one is visibly worse than the reduced one.
--
This is an art archive, so logically speaking, we want to archive the original vision, idea, and result of the artist that worked on it, you wouldn't want to store in a museum an attempt of copying the Monna Lisa and say it's the real thing, when the original is still available, and being preserved.

Knew it was Mairo before even clicking. Although not quite as extreme as your example, I had something similar happen to me with post #3651303.

watsit said:
Compression and resolution are both considered.

Again, I don't disagree; a 1300x1300 compressed JPG should be deleted in favor of a 1250x1250 PNG. The issue is when large increases in size are thrown away because the much smaller version has no artifacts. Yes, compression lowers the quality of an image's details, but it shouldn't be replaced with an image where those details are barely even visible. The details of the hatching of the eyes and the darker shadows, for example, are obscured in the lossless Tumblr versions of the posts I provided because they are so much smaller.

watsit said:
Twitter in particular also has a habit of resizing images to have a 2048 or 4096 width or height, which causes it to further deviate from the original on top of recompressing it as a lossy JPG.

The originals in question are inaccessible, so comparing them to any of the example posts serves no purpose.

m3g4p0n1 said:
Here's an original (sfw) drawing of mine, featuring 3 versions.

...

If you open all three images on individual tabs, and switch over them, you will noticed that the twitter one is visibly worse than the reduced one.

Your examples don't work because they are all almost exactly the same resolution, in which case I absolutely agree the compressed version should be deleted. For an equivalent comparison, however, imagine if your original file was actually 2200x1242 and the Twitter version retained this size, while your other two files were both samples (images where the website you uploaded them to has reduced their size). Which version should be kept then? The answer is the larger one, because despite Twitter's compression there is still far more information and detail in the larger version than the lossless smaller versions. Compressed details are better than lost details.

zapdos said:
[...]

Oh, okay, I see your point now.

https://i.imgur.com/OH1kWbY.png

And it reminds me of this example post #3801821

Twitter one is a 3500 x 4000 jpeg, while FA is 1795 x 2052 png.
Mairo explained in the comments why the FA one should be the chosen one though.

From what I understand from it, the priority is to have non-edited artwork, always. That also means no compression unless it's the only option, since twitter compression is a third-party edit, technically speaking.

FA's new image upload size just infuriates me, with their arbitrary 2K resolution limit rather than just enforcing a filesize limit like every other website with a sane leadership would do. I've seen FA's files frequently get bigger when they downscale them compared to the original file.

When will artists just start uploading to galleries that aren't awful and butcher their images? "Larger JPG/smaller PNG" always going to be a source of contention until people just start using a better website.

m3g4p0n1 said:
From what I understand from it, the priority is to have non-edited artwork, always. That also means no compression unless it's the only option, since twitter compression is a third-party edit, technically speaking.

The FA one is likewise edited - the original is presumably a 3500x4000 png and it was downscaled.

Twitter chooses to convert to jpg and Fur Affinity rescales the file. Neither is the original file.

m3g4p0n1 said:
Mairo explained in the comments why the FA one should be the chosen one though.

No, Mairo explained in the comments why someone with a $5000 professionally calibrated monitor might prefer to look at the PNG instead of the JPG. This is a particularly awful example of choosing to keep the inferior version and one of the worst takes I've ever seen on size vs compression for the purposes of imageboard archives, as it doesn't hold any water when you consider one extremely simple and very obvious fact - literally everyone can see a difference in size (especially when it's this drastic), while many people aren't even aware of what artifacting is (let alone chroma subsampling and macroblocking). A majority of the userbase of the internet in general would look at this and be rightfully confused as to why the bigger and better post was discarded. When someone in charge of deleting duplicates is saying an image that's literally double the size only has "ever so slightly more pixel detail", it really makes you wonder how much damage they've already done behind the scenes.

zapdos said:
I recently flagged post #1148023 and post #1148022 for being inferior to post #3897885 and post #3897810 respectively (they are almost 2/3 times bigger), but the much smaller Tumblr samples ended up deleted instead because the Twitter versions had some compression artifacts on account of them being from Twitter. I was told by the Janitor who deleted them that "[e621 does] not go by resolution" when choosing which version to keep, but this is directly at odds with the flag page which lists being larger as a reason to keep one version over another.

When I asked, the same Janitor also disclosed that they might keep even a 500x500 PNG from Pixiv over a 3000x3000 Twitter JPG, which is just absurd. Being over 60% bigger should already be enough to offset such compression, but apparently even if those images were 500% bigger they still wouldn't have made the cut. I could understand if the image was only, say, 10% bigger, or if the compression was akin to some deep-fried meme, but completely ordinary run-of-the-mill Twitter compression isn't nearly enough justification to ignore such a substantial size increase.

The flag for deletion wiki page and the "Duplicate or inferior version of another post" option on the New Flag page need to go into more detail in regards to what actually constitutes as an inferior version, the staff needs to be consistent with said rules when selecting which post to keep, and substantially larger images should not be thrown out for mere Twitter compression.

The difference is minimal. The second artwork that was deleted was ~2100 pixels tall meanwhile the one that was kept is 1280 pixels tall. Your analogy is off the rails, I am sure the janitor would prioritize the image that is 3000 pixels big over the one that is 500 pixels big with or without compression, but the difference is negligible enough that the compression artifacts were a bigger problem to worry about.

zapdos said:
No, Mairo explained in the comments why someone with a $5000 professionally calibrated monitor might prefer to look at the PNG instead of the JPG. This is a particularly awful example of choosing to keep the inferior version and one of the worst takes I've ever seen on size vs compression for the purposes of imageboard archives, as it doesn't hold any water when you consider one extremely simple and very obvious fact - literally everyone can see a difference in size (especially when it's this drastic), while many people aren't even aware of what artifacting is (let alone chroma subsampling and macroblocking).

Doesn't change the fact that the artifacting exists, and you don't need a high quality calibrated monitor to notice. It's one of those things where once you realize it's there, it becomes difficult to not notice. A PNG is much cleaner and more representative of the original image despite being resized, whereas Twitter may or may not have resized it anyway on top of adding JPG artifacts. While I might disagree over "keep even the 500x500 PNG from Pixiv over a 3000x3000 Twitter JPG" (it would make more sense to me to keep both with such an extreme disparity), PNG should be preferred over JPG. You can do more with a clean PNG (including upscaling it with various modern upscalers) than a noisy JPG (which is difficult to process without the noise interfering).

zapdos said:
A majority of the userbase of the internet in general would look at this and be rightfully confused as to why the bigger and better post was discarded. When someone in charge of deleting duplicates is saying an image that's literally double the size only has "ever so slightly more pixel detail", it really makes you wonder how much damage they've already done behind the scenes.

The site archives what it considers to be the best version to archive, and users can look at it as they wish. The majority of the userbase doesn't need to understand why one is chosen over the other, since as you mention, most people aren't even aware of these issues. Obligatory XKCD.

NotMeNotYou wrote this some time ago because this kind of topic keeps cropping up constantly: e621:image_quality

faucet said:
FA's new image upload size just infuriates me, with their arbitrary 2K resolution limit rather than just enforcing a filesize limit like every other website with a sane leadership would do. I've seen FA's files frequently get bigger when they downscale them compared to the original file.

When will artists just start uploading to galleries that aren't awful and butcher their images? "Larger JPG/smaller PNG" always going to be a source of contention until people just start using a better website.

This is the basic summary of it.
In past, artists who cared simply reuploaded the original quality version and those who didn't care, twitter had such massive resolution increase that it was easy to just go with twitter version. This wasn't ideal situation either, but it was pretty easy to enforce.

Nowdays because of FA management being absolutely fucking retarted, artists who want to upload original quality for their viewers have to pay (instead of users having to pay, like they do on places like Patreon, not sure does Pornhub still lock 1080p behind subscription?) and still won't be able to upload original quality because there's also 8.3MP cap even for paying users!
This is also why Twitter is similarly stupid as they similarly lock 1080p video quality behind Twitter Blue, meaning animators have to pay to upload higher quality for their viewers, then it's forced even for mobile viewers who might actually benefit from seeing 720p stream instead but there's not even a toggle for viewers for this!

Meanwhile sensible sites (read: basically all other gallery sites) do sampling, offer that lower quality sample when simply browsing, viewing or being previewed on messaging app and then full quality is toggle, seperate download button or simply kept in safe on server either for site or artist. Actual limits for artists are in form of filesizes because that what matters the most for actually storage costs and then sampling can be adjusted according to bandwidth costs.
This excludes social media as those just sample, compress and throw original in trash.

But basically what FA has done is worst of both worlds, they now have to use more bandwidth, they now have to store more and nobody is really happy and because artists still only use FA and Twitter, we are now in this headache inducing situation between Dragoneer and Elon Musk.

As for files hosted on e621, I still prefer lossless downscaled media over compressed one.
Justification I have for this that even though higher resolution can be perceived higher quality one, problem is that noise that JPG compression has on image, can be falsely seen as being more detailed (this is why many artists deliberately apply noise on their images at the post processing phase), even if that noise is literally from compression ruining detail, rather than being actual detail. If you start to denoise compressed image, even if that noise was detail it's now lost as well as compression artifacts, meanwhile scaling up lower resolution image might actually look better when it just tries to upkeep detail that's almost certainly correct detail.
Only problems are when resolution difference gets higher and smaller pixel level high contrast detail amount is higher, at which point that detail gets completely lost in lower resolution version, even if lossless.

Also remember that with Twitter specifically, compression rate is 85 and the JPGs are also chroma subsampled to YUV420, meaning that color information is also halved in dimensions.
With these specific examples, the images from twitter have 2048 in dimension, meaning they are most likely not original resolution either, so they are also downscaled from original resolution, meaning twitter downscalings interpolation still causing quality degradation as well. Color information in these files specicially is 768x1024 which is lower from 959x1280, so even if luma is higher resolution and human eyes perceive that better, half of the resolution is actually much lower, so even the resolution argumentation isn't as black and white as "this resolution high, other low".

Also fallback janitors have in situations they are unable to objectively or justifyabely resolve, is that the uploads are deemed identical and first upload is kept. We have 10 source fields so rest can be filled in. Also in these cases, artists should start uploading on places that do not alter their uploads and uploaders are technically free to go and bother artists for original copies.

wat8548 said:
Knew it was Mairo before even clicking. Although not quite as extreme as your example, I had something similar happen to me with post #3651303.

Yeah, in that case the resolution difference is pretty low and there's a lot of really subtle detail that's just mushed into macroblocks, especially againts that colored background that I would never take that JPG over PNG.
...and that's another thing, if it's super colored image with general detail then PNG is always better, but if it is extremely noisy grayscale image, now the JPG might be worth it.
It becomes "it depents" situation that's dependant on the actual content that's in the file which makes this even more of a headache when users ask "furaffinity or twitter" beause "IT DEPENTS™️" and it depents on both actual content and technical details.