hmmm...how to set parent child on existing posts??...is that a hidden interface?

tag:gun
tag:gun_gale_online
tag:ochiai_hitomi
tag:sinon
tag:sword_art_online

You can set this as parent child by clicking the option "Make this post the parent" You hae to enable "Advanced Editing" on your user options.
Wait... why this is 68mb

why this is 68mb

5487x8009 24bit png ?_?

I really get tired of your oversized/overfiltered scans..

@fireattack: How do you come to the conclusion that it is overfiltered?

Just flipping back and forth between the jpg browser resized-image the site uses for the solo-image's display page in 2 separate tabs you can clearly see more detail in the newer image. So how can you say it is overfiltered?

As for "over-sized": the source is a ~17x22 inch wall calendar, which takes a bit of work to scan in, but can allow for a lower DPI at scan time that can result in a lower noise level -- resulting in cleaner than average scan. Perhaps that's why you think it was "over-filtered"?

@booregardo: 5488x8009x32-bit (rgba) =~ 168MB. So I guess compression would be why the png was only 68MB. :-|

High resolution is great. I can even see the stroke.

@Spyps -- um... "stroke"?
thanks...

Brush strokes. A good quality scan will show the analogue/digital brush stroke.

@Astara

This image is strongly oversharpened, and there is a large amount of color noise and bright/dark halo artifacts everywhere, particularly around the edges and linework. Some of the fainter linework in the hair is also broken and partially missing.

Overall, all these things combined are what killed compressibility on this image and resulted in the large filesize. I did a quick test denoising and removing some of the overshapening and filesize came out at ~36MB which seems slightly more reasonable at nearly half the file size.

General policy for yande.re, is scans need to be fully denoised with smooth colors & gradients where applicable, important detail/lines/edges preserved, and generally just look clean and pristine when viewed at original 100% size. This one really does not look decent until it's scaled way down to hide the color noise and sharpening artifacts.

If you look at drop's scan, it actually has more detail in some areas than this one, despite being softer. Though if for example you clean, strongly sharpen, and color correct drop's scan to look similar to yours, you end up with something like Post #306518 which has a noticeable more detail in the grass, and without the lineart issues. Granted it seems to be a different source, but this kind of clean appearance without notable sharpening halos is generally what you should aim for when filtering.

Cyberbeing said:
Granted it seems to be a different source, but this kind of clean appearance without notable sharpening halos is generally what you should aim for when filtering.

I'm not sure, but you're aware that you totally oversharpened that?

_Astara said:
@fireattack: How do you come to the conclusion that it is overfiltered?

By over filtered I mean over sharpened (as cyberbeing already mentioned), not over smoothing/denoising. And I didn't mean the child scan is better than yours if I made this confusion.

Schezza said:
I'm not sure, but you're aware that you totally oversharpened that?

Yes, I'm totally aware which is why that post is held and hidden. Normally I don't sharpen things to even 10% of that level.

That post was only a quick example to show the absolute limits of what is possible when sharpening without massive halos or artifacts. The goal was to match the general oversharp appearance, contrast, and color of Astara's post in order to make the differences in details & linework to drop's scan more easily identified. I felt I needed to do this for comparison purposes, since Astara seemed to be confusing oversharpening artifacts and noise with actual detail present in the scans.

The key difference to take note of, is that drop's original scan Post #289101 retained detail, despite being filtered smooth with a considerably different color tone. That type of smooth filtering without detail loss is more in line with expectations when uploading scans here. Sharpening in general should be optional, and only performed to recover minor sharpness lost when denoising, not further enhance the source material to an extreme like this post or my example.

I didn't apply a separate sharpness filter. The only sharpness applied was recovery sharpness in the filter itself.

Looking at your image though -- usually I work with the colors to color balance it more, but in this case, both the red and blue histograms were cut off in the low end (true in the raw scan as well), while the green histogram was shifted about 10-15% toward the right end (compared with the red and blue ones).

Also, I looked for the best balance in the stonework that was near Sinon that framed her as well as the look in her eyes. I'll admit to not paying as much attention to the grass -- but looking at the original, it may have been upscaled, smoothed and sharpened by the publisher. As for breaks in the line art in her hair. That's true of the original (as printed by the publisher).

Given the less attention to detail in the grass, I'd say the publisher (or the original artist) did corrections to accommodate the larger print size -- i.e. it looks like they did correction on the center pieces, but not as much in the grassy areas.

FWIW --- I'm not submitting this as a 'raw' scan. It is not. I did noise removal on it noise ninja (couldn't get as good a result with GREYCstoration -- but am not as familiar with it either).

But the calendar was purchased through J-List, and just came -- perhaps someone else wants to buy a copy and to show how it should look? It seems maybe a global sharpening was done by the artist or their publisher and the areas in the grass, to my eye, look like they've had a manual or spot-blur done to them, perhaps to focus more on the center subject? Note, these are retrospective comments -- not something I thought about before posting -- as I don't have another source for this to compare it against.

_Astara said:
I didn't apply a separate sharpness filter. The only sharpness applied was recovery sharpness in the filter itself.

Then your denoise filter itself is to blame. Noiseninja, NeatImage, Topaz, and other such paid Photoshop filters almost always oversharpen with default settings. In general, you should disable sharpening entirely in such denoise filters, since they are tuned for photographs rather than anime images. With Noiseninja, this means setting USM (unsharp mask) to 0.

From your last few uploads, I think you should just avoid doing any sharpening at all on future uploads for the time being and work instead on smooth denoising without major detail loss. If you still find it soft, you could always resize it down to 300-350dpi afterwords to regain some clarity + anti-alaising. Most of your uploads seem to be severely underfiltered on high-frequency color noise, and somewhat overfiltered on medium/low-frequency detail. The last thing you want to do is enhance noise and artifacts from your denoising, as you seem to be doing apparently by accident via Noiseninja.

I'd highly suggest taking some time to learn and improve your skill at denoising in a balanced manner. While paid Photoshop denoise filters can sometime be useful for evening out and reducing very strong noise, they often do more harm than good on anime artwork when used as your primary method of filtering. GreyC is your best friend, and your most powerful tool for restoring solid colors and gradients when used with proper settings. Like anything else though, it requires some practice and general willingness to experiment and learn.

_Astara said:
Looking at your image though -- usually I work with the colors to color balance it more, but in this case, both the red and blue histograms were cut off in the low end (true in the raw scan as well), while the green histogram was shifted about 10-15% toward the right end (compared with the red and blue ones).

Don't worry much about this, since that wasn't really the focus of my example at all. I was only attempting to somewhat match the appearance to your upload. Any differences in the color balance and contrast more have to do with drop's scans looking rather different overall. Since you are actually scanning your own source material, just attempt to match it visually under a good light. Preferably one of those common 'Daylight' 6500K CFL bulbs to match the D65 color temperature of a calibrated monitor.

_Astara said:
As for breaks in the line art in her hair. That's true of the original (as printed by the publisher).

Are you positive the lines don't just have very faint sections like drop's scan, rather than actual breaks in the lines?

Take a look at 400% crop from your scan in the hair area, and what do you see? Blocking, massive halos, color noise, gaps in the hair, and other strange artifacts: http://i.imgbox.com/QQJ9MOks.png

You need to pay more attention to avoiding and eliminating issues such at that, even if they exist in the original source. For comparison this is what drop's original scan looks like. Some jpeg compression artifacts, but otherwise clean: http://i.imgbox.com/P5qSGnNQ.png

@Cyberbeing. Above you said the compare limit is 100% of the scan. That's *mostly* fair, you are saying above that I need to go to 400% to see the defects??

Um... Looking at a scan that was designed for someone working down in size @ 400% to see defects is a bit harsh, no? If this was at "screen size" resolution such that someone using it to create a new image, couldn't use a smaller size, then maybe, but I always *try* to work in sizes of at least 2.5-4.5x of the final work. If looking for scratches,

That said, I agree in that it had had more sharpening artifacts than I thought were in there... will have to see what I should do w/that.

As far as drop-outs in the hair -- you weren't specific as to where, but the lines are pretty rough looking -- especially @ 400% ( ;-/) ((NOTE: saw your image of area to compare, so updated image.

Now let's see if this works (i.e. raw hair area so you can compare against the raw of this)


Ah... I think it might work now?... still not seeing image inline in comments...just a blank area... but at least I see the image on the inline page?...

If you can see it, that area is pretty sketchy... but that might be the over-sharpening you mentioned... Have another version to try -- but needs to have scratches & dust cleaned.. but looks like it got the pattern w/o the haloing...

Have to find time to re-remove the scratches though...(sigh).
It's about the same in size though. If I run it through a Gaussian blur, that'd probably bring it down considerably.

_Astara said:
@Cyberbeing. Above you said the compare limit is 100% of the scan. That's *mostly* fair, you are saying above that I need to go to 400% to see the defects??

No, the oversharpening and luma blocking is clearly visible at 100% as well: http://i.imgbox.com/JSaPgvfS.png

Yes, the high frequency color noise and luma patterns in the flat areas are somewhat less obvious, but at least on this scan, both are trivial to remove at such low intensities. Along with a minor visual benefit, removing high frequency noise should improve png compressibility considerably so it's a good habit to have as long as it doesn't cause severe banding.

For the color noise: Photoshop's default 'Reduce Noise' filter, 'Strength' 0 + 'Reduce Color Noise' 20%-25% + 'Sharpen 0%' is almost always sufficient. Though since you own NoiseNinja, you may as well use that, just remember to disable USM and set Luminance strength to 0. Simple.

For the luma patterns: Similarly, any luma denoise filter at low strength should do it. Anything dealing with luma is harder to generalize though. Your other scan Post #305791 is a better example of a problem with luma patterns, and would require something a bit stronger. Normally I'd use GreyC for problems such as this, but that always requires some fine tweaking to make it do exactly what you want.

There is nothing wrong with filtering images in multiple steps at different strengths and different filters. I'd even recommend you always perform filtering on a duplicate layer, which allows you the flexibility of blending at a lower opacity if desired, as well as easier comparisons than messing with history states.

I will say though, it's not a bad idea to occasionally check your intermediate filtering results at higher zoom, since at least for myself it saves time in the long run. I can't tell you how many times I've forgotten to do this, only to discover an hour later that I messed something up early on, and now have to start over from scratch. At the very least, you should give the image one in-depth checkover as the last step before uploading.

p.s. Your raw crop doesn't look to bad. There isn't any sharpening or noticeable luma artifacts around the edges, the line is just very very faint which makes it tricky to not filter it to oblivion when denoising. Something like following cleans it up the crop decently, but you'd need to watch out for those GrayC settings being overly destructive to fine detail elsewhere:
NoiseNinja: http://i.imgbox.com/jNFVjdCC.png
+
GreyC: http://i.imgbox.com/qibLFLEx.png
=
Result: http://i.imgbox.com/ORFf5mQ3.png
http://screenshotcomparison.com/comparison/105239

I posted a new scan that I am pretty sure you'll like better given the similarity of the smoothing in the raw crop. In some ways it is not as clear as the first, but the first is maybe 'clearer than it ought to be'...I think I restored some of the detail in the grassy area. The green histogram was overpowering the red and blue and in the new version the face doesn't look so washed out (too bright).

The stone work looks clear enough now for me to guess the artist used reality sampling like many anime's seem to be doing these days....Anyway, hope it looks better? I don't see the halo's in the hair, but I have similar detail in the area you showed for the broken lines..