Hello guys, I just saw this post #86879 getting deleted for both "poor use of AI upscaling" and "compression". I'm pretty confused over that claim and don't understand what's there to fix cuz to me the picture is pretty clear and I don't see what artifacts might be caused by upscaling so if anyone can help me identify them that would be great.

The picture: https://drive.google.com/drive/folders/1XiArS8BJGhj-Ukchvwi-kNo7LLxiNI6A?usp=sharing
(fyi there's a depth of field on this image, if you zoom in you can literally see the "canvas" effect of this piece so I particularly don't get what they mean about "compression")

Again I'am completely willing to fix obvious problems but this feels like there's nitpicking for details here.

meara said:
Hello guys, I just saw this post #86879 getting deleted for both "poor use of AI upscaling" and "compression". I'm pretty confused over that claim and don't understand what's there to fix cuz to me the picture is pretty clear and I don't see what artifacts might be caused by upscaling so if anyone can help me identify them that would be great.

The picture: https://drive.google.com/drive/folders/1XiArS8BJGhj-Ukchvwi-kNo7LLxiNI6A?usp=sharing
(fyi there's a depth of field on this image, if you zoom in you can literally see the "canvas" effect of this piece so I particularly don't get what they mean about "compression")

Again I'am completely willing to fix obvious problems but this feels like there's nitpicking for details here.

I’m curious as well. Don’t see anything obvious.

Hello!

I just wanted to drop by to communicate my thoughts on why I made the decision I made on https://e6ai.net/posts/86879

So I am taking another look at the image again and for me what influenced my decision primarily was the graininess that you can see most evidently around the following parts of the image (here is a link of your piece with the parts highlighted https://gofile.io/d/iWnmIL):
-bottom part of dress where black meets white
-eyes
-viewer-right front paw
-viewer-left front paw (slight toe invisibility occurring here also)

As a catch-all, I marked it as an upscaling/compression issue. The upscaling I marked by habit because it's a large image; I'm not certain if this is an issue from the mode/checkpoint following a large upscale around 2x or possibly beyond. The compression I marked because, well, I see the graininess and in my opinion for a piece that has focus on specific parts of an individual's body, that stands out. Now potentially this is explained because she is not in focus but her private area is. To that I would still have to point out the issue with part of her left hand going invisible. I suppose the graininess around the rest of the image is too distracting.

I apologize if this comes across as nitpicking, but I am just walking you through what I saw and what influenced my decision.

asscas said:
Hello!

I just wanted to drop by to communicate my thoughts on why I made the decision I made on https://e6ai.net/posts/86879

So I am taking another look at the image again and for me what influenced my decision primarily was the graininess that you can see most evidently around the following parts of the image (here is a link of your piece with the parts highlighted https://gofile.io/d/iWnmIL):
-bottom part of dress where black meets white
-eyes
-viewer-right front paw
-viewer-left front paw (slight toe invisibility occurring here also)

As a catch-all, I marked it as an upscaling/compression issue. The upscaling I marked by habit because it's a large image; I'm not certain if this is an issue from the mode/checkpoint following a large upscale around 2x or possibly beyond. The compression I marked because, well, I see the graininess and in my opinion for a piece that has focus on specific parts of an individual's body, that stands out. Now potentially this is explained because she is not in focus but her private area is. To that I would still have to point out the issue with part of her left hand going invisible. I suppose the graininess around the rest of the image is too distracting.

I apologize if this comes across as nitpicking, but I am just walking you through what I saw and what influenced my decision.

I know it’s not my submission, but I’m honestly trying to see what you’re looking at as I’m going to be inquiring about all of mine that were deleted recently. I’m at a loss for words.

asscas said:
Hello!

I just wanted to drop by to communicate my thoughts on why I made the decision I made on https://e6ai.net/posts/86879

So I am taking another look at the image again and for me what influenced my decision primarily was the graininess that you can see most evidently around the following parts of the image (here is a link of your piece with the parts highlighted https://gofile.io/d/iWnmIL):
-bottom part of dress where black meets white
-eyes
-viewer-right front paw
-viewer-left front paw (slight toe invisibility occurring here also)

As a catch-all, I marked it as an upscaling/compression issue. The upscaling I marked by habit because it's a large image; I'm not certain if this is an issue from the mode/checkpoint following a large upscale around 2x or possibly beyond. The compression I marked because, well, I see the graininess and in my opinion for a piece that has focus on specific parts of an individual's body, that stands out. Now potentially this is explained because she is not in focus but her private area is. To that I would still have to point out the issue with part of her left hand going invisible. I suppose the graininess around the rest of the image is too distracting.

I apologize if this comes across as nitpicking, but I am just walking you through what I saw and what influenced my decision.

I'm sorry but that decision of yours is quite severe for such minute details imho, the grain wasn't supposed to be seen as an issue and I chose to keep it to give that "paint on a canvas" effect to the piece and you can't even really see it unless you zoom in pretty close...
I understand the tightening of the rules regarding quality and I've honestly been having fun fixing and editing my outputs since then but, respectfully, the quality requirement is ridiculous this time.

crashbandit said:
I know it’s not my submission, but I’m honestly trying to see what you’re looking at as I’m going to be inquiring about all of mine that were deleted recently. I’m at a loss for words.

meara said:
quite severe for such minute details imho

The anatomy is a fundamental component of the current guidelines.

meara said:
you can't even really see it unless you zoom in pretty close...

All staff views images in full size.

meara said:
I understand the tightening of the rules regarding quality and I've honestly been having fun fixing and editing my outputs since then but, respectfully, the quality requirement is ridiculous this time.

Again I am really sorry if you disagree with my decision. Respectfully, I disagree and would also point out that I am not the sole janny on staff that feels the same; this was not an arbitrary or malicious decision.

My quality standards are pretty high, but I think in this case janitors are way too nitpicky. If these janitors' standards apllied to human-made art, over 99% of it would be deleted.

tyto4tme4l said:
My quality standards are pretty high, but I think in this case janitors are way too nitpicky. If these janitors' standards apllied to human-made art, over 99% of it would be deleted.

there is no reason for machine made art to be "low quality"
machines are supposed to be like
perfect and more accurate and faster, they shouldn't do mistakes like humans do, or take as much time as one does
literally like i said there is no reason to AI to be low quality anymore
if it is then you are either not doing it right or something else idk
never done AI gens myself so i dont know the process but shouldn't be that complicated

the first few posts were low quality because it was a start, now they are great and can always be great and will always remain great, thats the quality standards on how i see it
technology advances and the quality floor should get higher
like nobody wants a nokia or some old phone that isn't as in much quality as an xiaomi for example
if the better can always exists, no reason for "worser" ones to be on display
at least in my views but like i said
dunno the process and can't care myself to learn

Hello, the issues I had were more about the wrongly positioned anus, and the tail looking like it's detached. Anyway, while I found the result could be passable, Asscas decided that it’s not.

Instead of focusing on finding the bare minimum for what’s considered acceptable, maybe the energy could go toward actually improving the work so it doesn’t come close to that line in the first place. That’s something some of you involved here should think about.

I’ve now wasted about 20 minutes of my time, easily 10x longer than the original rawgen seemed to take, just to fix it. I did this because this feels like a better use of my time than arguing with you, how exactly lowest quality of acceptable image should look like.

https://postimg.cc/PpBxPK2k

Here you go. It's yours now.

All you need to do is start by smudging out the mistakes, maybe adjust things a bit with the warp tool. Then throw it into IMG2IMG, enable both settings on 'NEVER OOM INTEGRATED', and generate it in 2K or even higher if you want. Do it multiple times at different Denoise levels and then combine the best parts from all the outputs. That is all.

flowersylveon said:
the first few posts were low quality because it was a start, now they are great and can always be great and will always remain great, thats the quality standards on how i see it
technology advances and the quality floor should get higher

For consideration, even the very oldest, first images on e6ai - many of which look like ass - were technically capable of being pristine and error-free. It's just a matter of how much effort you were willing to put into it beyond the initial gen via the use of inpainting, which has been around since day 1.

This level of effort has been reduced through improved models over the years, meaning it's easier to successfully generate fixes without needing to RNG over and over again. That said, economic gatekeeping is a thing to consider - some of those technical improvements you speak of come at a cost of steeper hardware requirements not everyone may have.

edit: to clarify this post implies nothing to the OP's issue, I'm just commenting on your tangent there is all.

terraraptor said:
For consideration, even the very oldest, first images on e6ai - many of which look like ass - were technically capable of being pristine and error-free. It's just a matter of how much effort you were willing to put into it beyond the initial gen via the use of inpainting, which has been around since day 1.

This level of effort has been reduced through improved models over the years, meaning it's easier to successfully generate fixes without needing to RNG over and over again. That said, economic gatekeeping is a thing to consider - some of those technical improvements you speak of come at a cost of steeper hardware requirements not everyone may have.

edit: to clarify this post implies nothing to the OP's issue, I'm just commenting on your tangent there is all.

Hardware isn’t all that essential for SDXL. I myself am running one of the worst performers, a GTX 1070, which you can pick up for around €100, and I’m doing just fine without feeling the need to upgrade. At the end of the day every hobby, whether it’s sports or creative work, comes with some gear requirements. No one’s going to cut you any slack because you decided to buy a €5 tennis racket that’s barely holding together off Gumtree. The less you spend the more work you will have to invest yourself, that applies even to image generation. With my equipment, I should be the one crying here for equality.

flowersylveon said:
[...]
perfect and more accurate and faster, they shouldn't do mistakes like humans do, or take as much time as one does
literally like i said there is no reason to AI to be low quality anymore
[...]

That’s what many people believe, perfect machines. But sadly it is wrong. The AI learned from human made works. And it can’t tell what is a mistake or errors. Artists have a hard time drawing good hands, and so does the AI. It just will learn ALL patterns, good or bad. Also a huge problem with LLM that will give you biased responses…

never done AI gens myself so i dont know the process but shouldn't be that complicated

Getting high quality gens with almost no weird errors is harder than it seems. The longer you work on one image the more stuff that seems odd you will notice. Stuff that you didn’t notice when you looked for first time at the rawgen.
It is harder than you might think, maybe you should try for yourself. It can easily consume several hours to fix all the blatant errors in one image.

denatural said:
Hardware isn’t all that essential for SDXL. I myself am running one of the worst performers, a GTX 1070, which you can pick up for around €100, and I’m doing just fine without feeling the need to upgrade. At the end of the day every hobby, whether it’s sports or creative work, comes with some gear requirements. No one’s going to cut you any slack because you decided to buy a €5 tennis racket that’s barely holding together off Gumtree. The less you spend the more work you will have to invest yourself, that applies even to image generation. With my equipment, I should be the one crying here for equality.

Sure I getchya. I was remarking on the sentiment being thrown around that made it sound like the whole AI-using community is wholesale able to take advantage of the newest tech, "no reason not to". I'm thinking more long-term here, just looking at the stuff coming down the pipe. Even now, I can't even use flux on my own rig. Wish I could. And just wait until we get deeper into the world of AI animations, oh boy.

And fwiw, traditional artists can get their stuff permitted on e621 for like, what, $1? Pencil sketches are allowed, after all.

Anyway don't mistake my comment as being in favor of reducing quality control here. Being stuck on SD1.5 up until just a few weeks ago and having to spend upwards to 500+ inpaint iterations for hours and hours to get 1 image done, the whole "i dont know the process but shouldn't be that complicated" rubbed me the wrong way.

denatural said:
Hello, the issues I had were more about the wrongly positioned anus, and the tail looking like it's detached. Anyway, while I found the result could be passable, Asscas decided that it’s not.

Instead of focusing on finding the bare minimum for what’s considered acceptable, maybe the energy could go toward actually improving the work so it doesn’t come close to that line in the first place. That’s something some of you involved here should think about.

I’ve now wasted about 20 minutes of my time, easily 10x longer than the original rawgen seemed to take, just to fix it. I did this because this feels like a better use of my time than arguing with you, how exactly lowest quality of acceptable image should look like.

https://postimg.cc/PpBxPK2k

Here you go. It's yours now.

All you need to do is start by smudging out the mistakes, maybe adjust things a bit with the warp tool. Then throw it into IMG2IMG, enable both settings on 'NEVER OOM INTEGRATED', and generate it in 2K or even higher if you want. Do it multiple times at different Denoise levels and then combine the best parts from all the outputs. That is all.

A couple things and allow me to be rude:
1 - I spend a couple hours per day doing AI stuff for fun and I now almost always edit the results in photoshop when I'm confident I can fix it, sometime I do put it back in Img2Img to make it really part of the generation, I do NOT make low effort posts that take five minute from start to finish.

2 - I didn't argue anymore on this subject cuz I was "fine" with the final decision so why so damn rude all of a sudden?

3 - I didn't ask you to fix it nor did I REQUEST your opinion personally, so shove your precious wasted 20 minutes where to sun doesn't shine, you did it yourself out of your own free will.

4 - I didn't make this post solely to complain, I had already asked Asscas about another post that had been deleted and didn't want to ask them again since I'm guessing janitors are busy, so I just asked here if anyone could either tell me if this decision was maybe too harsh and/or which artifacts I should look for that could be caused by the described issues.

PS: regarding your fix while I like that the angle of the anus is better and that the DOF is weaker, I'm gonna be as nitpicky as you guys and refuse it, there's some weird artifact in her pussy, the claws on the right look worse and her hand is blurry. there's a strange shit smudge on her face and while I get your criticism about the tail looking a little disconnected because the part bellow her butt is pitch black your solution was making it all blurry? also there's a slight yellow tint in parts of her fur which wasn't there.

PPS:I don't even now what FlowerSylveon is even on about, the software ABSOLUTELY do make mistake which is why there's quality control on this website, if you don't know about the process and don't even care to know, don't come here argue about "the strength and certainty of steel".

meara said:
PS: regarding your fix while I like that the angle of the anus is better and that the DOF is weaker, I'm gonna be as nitpicky as you guys and refuse it, there's some weird artifact in her pussy, the claws on the right look worse and her hand is blurry. there's a strange shit smudge on her face and while I get your criticism about the tail looking a little disconnected because the part bellow her butt is pitch black your solution was making it all blurry? also there's a slight yellow tint in parts of her fur which wasn't there.

Yeah, I actually like your picture better than the "fixed" version, which also has a discolored part near the left edge (near the paw), probably due to inpainting.

meara said:
you can't even really see it unless you zoom in pretty close...

Here's the rational on upscaling: if it looks terrible at 1:1 zoom, why the fuck did you make it that big? The point of larger image sizes is to preserve fine details and retain crispness. AI upscaling tends to do the opposite: introduce merging features, blurring edges, adding artifacts, and so on.

No one wants to save an image that looks like ass when viewed that large. It's a waste of hard drive space. If it looks good at a smaller resolution, upload that instead.

draco18s said:
Here's the rational on upscaling: if it looks terrible at 1:1 zoom, why the fuck did you make it that big? The point of larger image sizes is to preserve fine details and retain crispness. AI upscaling tends to do the opposite: introduce merging features, blurring edges, adding artifacts, and so on.

No one wants to save an image that looks like ass when viewed that large. It's a waste of hard drive space. If it looks good at a smaller resolution, upload that instead.

As I said I'm fine with the final verdict. I was a little annoyed cuz I did spent a little time on it and didn't believe the originally mentioned details where enough to delete it but that was it PERIOD, the discussion should have ended...

Now going as far as calling it "TERRIBLE" I think that it's just wrong , but that's just my dogshit opinion so...
FYI this is the non-upscalled version https://drive.google.com/drive/folders/1uOB5OW-IWIlrfomDQIMaINZhMi12JI22?usp=sharing I just wanted to make it crispier since I thought it was a little blurry when zoomed in.