Here are two images on this site:

post #4130479 post #4141553

They're the same picture, but the one on the left (3840x2160 JPG from Twitter) is higher res with compression artifacts and the other (1920x1080 PNG from other sources) has no visible artifacts. If you look at the background in both pictures you can see the rougher gradient on the wall in the JPG version.

Do we prioritize resolution or quality?

One of the other images in this series of posts (post #4141559) has a higher res image at its twitter source, but that has artifacts visible when viewed at actual-size. Do we still replace it?

Other topics talk about this, but a lot of them didn't reach a consensus.

ironically granted it's censored, the png post includes source link to pixiv, which is both large as twitter's and kept as original png, but the uploader chose the inferior baraaag as image source instead.
for the topic itself, i think this similar thread and the comment could cover things better than i do.

bitez said:
ironically, the png post includes source link to pixiv, which is both large as twitter's and kept as original png, but the uploader chose the inferior baraaag as image source instead.

the pixiv version is censored

civil_nitrogen said:
Do we prioritize resolution or quality?

We juggle it the best we can.

Normally, yes, the .png is preferred over a .jpg because .png is a lossless compressed format whereas .jpgs always add in some amount of compression. However, FA and Twitter added an unpleasant new twist to this by FA giving us good quality but small .pngs and Twitter giving us larger sized but inferior quality .jpgs. Which is better? It took us a little bit, but we settled on a general guideline of It Depends™.

Helpful, ain't it?

Currently, the rule of thumb is that the .png is better unless the .jpg is at least around twice its dimensions. Even then, we have to visually compare the two, by using the SmartUpscale add-on to disable blur when zooming in on an image. The smaller .png is zoomed as big as it can go while the .jpg is zoomed to appear as close to the same size as possible. The two are then visually inspected to see if the compression crust is too noticeable and if there's more original information kept in the .jpg (small details are frequently lost when a picture is scaled down). Frequently, the Twitter .jpg does look better than the FA .png on both counts, but it's not unusual for the .png to sometimes still look better despite being smaller. In the former case, we keep the .jpg and in the latter, we keep the .png.

The most ideal solution? Hope the picture is on a different gallery that allows big .pngs or pluck up the nerve to get it directly from the source.

bitez said:
ironically, the png post includes source link to pixiv, which is both large as twitter's and kept as original png, but the uploader chose the inferior baraaag as image source instead.

This is a case when sites other than FA give us the same problem. None of the offered sources are ideal. Twitter/X is the usual .jpg, Baraag is taking the place of FA with a smaller .png, and Pixiv is censored. The best version of the picture was probably the DA upload (found thanks to SauceNAO), but that's now a dead source. The uploader likely decided a non-censored though small .png was the best option to go with. That's not an unreasonable choice as e621 does tend to prefer uncensored versions when possible.

clawstripe said:

Currently, the rule of thumb is that the .png is better unless the .jpg is at least around twice its dimensions.

In our case, the .jpg is double the .png's resolution.

clawstripe said:

Even then, we have to visually compare the two, by using the SmartUpscale add-on to disable blur when zooming in on an image. The smaller .png is zoomed as big as it can go while the .jpg is zoomed to appear as close to the same size as possible. The two are then visually inspected to see if the compression crust is too noticeable and if there's more original information kept in the .jpg (small details are frequently lost when a picture is scaled down). Frequently, the Twitter .jpg does look better than the FA .png on both counts, but it's not unusual for the .png to sometimes still look better despite being smaller. In the former case, we keep the .jpg and in the latter, we keep the .png.

I don't have SmartUpscale, but I did look at both in full screen using my 1080p monitor (the .PNG's exact res). Flipping between the two with the .JPG at 50% zoom, the .PNG is superior (has more colors in the gradient). Zooming in at the JPG's 100% and PNG at 200%, the PNG is blurry (using Linear, Cubic and NoHalo interpolations) but the JPG's artifacts are visible, especially around the butt-crack lines.

My conclusion, the PNG is better. I would flag the JPG as Inferior/Duplicate. But I feel someone else who knows better, should make the call.

civil_nitrogen said:
In our case, the .jpg is double the .png's resolution.

By "double" the resolution, that actually means quadruple: double width and double height. But even then, that's only considering the 4:2:0 chroma subsampling, where JPGs have a quarter of the color resolution as luma/brightness resolution, but JPGs will typically have further degradation with noise around hard edges and color banding in gradients, with colors being slightly off. So unless it's a really high-quality JPG encoding that have larger filesizes to typical JPGs, and thus reducing the benefits offered by the format, I'd still consider a PNG superior to even a 4x JPG.

civil_nitrogen said:
They're the same picture, but the one on the left (3840x2160 JPG from Twitter)

I think it's was just upscaled version. Image with such simple draw style quite easily upscales via AI upscalers with sharp look. Also, AI upscaler quite easily deletes JPG artifacts.

Normally, if I were the uploader in this situation, I would just ask the artist for the original file to avoid all the hassle of choosing between compressed sources.

If you are intrested how it looks via AI upscaler I can upload it to someone. Looks similar with JPG version, but without artifacts.

yetanotheraiuser said:
If you are intrested how it looks via AI upscaler I can upload it to someone. Looks similar with JPG version, but without artifacts.

https://e621.net/help/uploading_guidelines#bad

Bad Things to Upload:

[...] artificial upscales, AI/neural network edits, etc.

pleaseletmein said:
the pixiv version is censored

admitedly forgot about that

clawstripe said:
This is a case when sites other than FA give us the same problem. None of the offered sources are ideal. Twitter/X is the usual .jpg, Baraag is taking the place of FA with a smaller .png, and Pixiv is censored. The best version of the picture was probably the DA upload (found thanks to SauceNAO), but that's now a dead source. The uploader likely decided a non-censored though small .png was the best option to go with. That's not an unreasonable choice as e621 does tend to prefer uncensored versions when possible.

true, i've also wanted to mention while i was sleep deprived about furaffinity's reduction to the maximum size target, especially leaving very indistinguishable artification.
in couple of circumstances if there is no uncensored nor uncompressed png images available, it has a high chance of censored version posts being approved together with a jpg post. e.g post #4752924

watsit said:

If it was not original source and just someone uploaded 4K version, than it may be AI upscaled version from 1920x1080. On my upscale that looks similar.

I saw some crazy stuff like someone took a damn screenshot with a cell phone where a PNG (or smaller JPG) was scaled to fit the screen, and it used the phone's native resolution.
And of course it not only looked worse at a higher resolution and filesize, but it also had some of the GUI elements and in a few cases, even clipped parts of the original image away.

I've also seen Twitter upscale images for no GD reason. Like, there was never a file that resolution uploaded to them.

I've been resorting to using IrfanView or other tools to do comparisons. I've seen PNGs with obvious JPEG artifacts on object edges. Sometimes the metadata is a dead giveaway. Resizes don't often mention Adobe products. ;)