Pixelated image when resizing layer

[javierr]

11 hours ago, debraspicher said:

I think we are talking about layer operations, not document-wide. Anyway I usually crop then do document resize on single image with my target proportions/dimensions. I can see when importing multiple layers for compositing and resizing, the blurriness can be rather annoying when viewing outside 100%. Especially if the full image doesn't fit in the viewport.

Yeah, my question it is basically about working with the program, manipulating layers, resizing them... (zoom is not my problem neither, as the layer is not clear at any zoom level once the layer is resized).

About exporting the file (with File>Export), I don't think there is an issue there, as it looks like is working fine, because it exports what you see in Photo Persona (being that blurry, good or bad, but is that, it respects what is in Photo Persona). You even have several options to try exporting -bilinear, lanczos, etc.- so the export part is fine.

About Export Persona, that is another issue pointed by @Chris B. It looks like is using a "vector view" and does not take the image directly from Photo Persona. It is corrected in macOS, but not in Windows.

Hope developers fix it quickly as I can not use Affinity Photo and have to keep using (and montly paying) Photoshop.

16 hours ago, debraspicher said:

I believe in PS, it was in preferences but not 100% sure

I attach an image of my Photoshop preferences (Bicubic automatic was set).

[javierr]

17 hours ago, debraspicher said:

Is this the issue you are having? I tried a 2nd try rasterizing before resizing. Still shows the blurriness like this screen you provided.

Yes, that is exactly the problem. I also tried rasterizing the image, but with no better results.

[NotMyFault]

2 hours ago, javierr said:

Well, if the bug is showing the image correctly in Export Persona, welcome that bug to Photo Persona! 😁

 

The export persona should give a as-realistic-as-possible preview of the exported result. 

In case of this bug, it doesn't: it shows more details (higher resolution) than the exported file will have.

There is absolutely no benefit.

If you prefer to have an unrealistic vector preview, you can use Designer in Vector View mode. Its self-deception.

One fundamental principle of Affinity Photo is that it tries (not always successfully) to give a preview of how the exported result will look, as raster / bitmap format.

Designer in contrast, has several view modes, and lets you choose if you want to get a virtually unlimited resolution using Vector view mode, or kind of similar to Affinity Photo raster / bitmap preview.

[javierr]

5 minutes ago, NotMyFault said:

The export persona should give a as-realistic-as-possible preview of the exported result. 

In case of this bug, it doesn't: it shows more details (higher resolution) than the exported file will have.

There is absolutely no benefit.

If you prefer to have an unrealistic vector preview, you can use Designer in Vector View mode. Its self-deception.

One fundamental principle of Affinity Photo is that it tries (not always successfully) to give a preview of how the exported result will look, as raster / bitmap format.

Designer in contrast, has several view modes, and lets you choose if you want to get a virtually unlimited resolution using Vector view mode, or kind of similar to Affinity Photo raster / bitmap preview.

It was a joke...

[SrPx]

It wouldn't be OK if a layer being transformed, is showing up more blurry than it should. IE, more blurry than those actual pixels are in reality, due to an internal (optimization for performance) engine thing. This is what I had understood that happens with their latest posts (Javier and Debra). Did I get it wrong now?

The way I see it, blur for optimization would be fine when if the hardware is low, and set to perform so in preferences. But would be ideal if we can "force" it in preferences to work 1:1 with pixels, no extra blur added. Again, I am trying to understand the issue, mostly.

EDIT: I mean,  not 1:1, as I guess in 100% zoom there's no problem. But with a good and sharp algo when working on a zoomed out level.

BTW, if I understood finally the problem, this used to happen in arcane/older Photoshop versions...

 

[NotMyFault]

2 minutes ago, javierr said:

It was a joke...

Sorry if my joke detector was out of calibration

[javierr]

12 minutes ago, SrPx said:

It wouldn't be OK if a layer being transformed, is showing up more blurry than it should. IE, more blurry than those actual pixels are in reality, due to an internal (optimization for performance) engine thing. This is what I had understood that happens with their latest posts (Javier and Debra). Did I get it wrong now?

Hi, @SrPx. Yes, that is the problem. Affinity Photo (in Photo Persona) it is not showing correctly a resized image layer. You can try the steps for yourself as described in my original question (first post).

 

13 minutes ago, SrPx said:

The way I see it, blur for optimization would be fine when if the hardware is low, and set to perform so in preferences. But would be ideal if we can "force" it in preferences to work 1:1 with pixels, no extra blur added. Again, I am trying to understand the issue, mostly.

I don't think Affinity is making a blur optimization because of low hardware or software, mainly because I have Photoshop installed in this same laptop and works correctly. I think there is a bug (which is strange so nooobody has pointed it in version 1 or 2) or Affinity is set to strange image screen perfomance and there is no place to change it.

 

20 minutes ago, SrPx said:

EDIT: I mean,  not 1:1, as I guess in 100% zoom there's no problem. But with a good and sharp algo when working on a zoomed out level.

In 100% zoom there is the same problem. Please try for your own with the steps described in my first post. In 100% you will see the image very small in screen, but still you can notice it is not sharp. Anyway, you can export it and check and zoom the exported file: it will be as blurry as you see in Affinity screen (unless you export it with lanczos, but that is a workaround, not a real solution).

 

23 minutes ago, SrPx said:

BTW, if I understood finally the problem, this used to happen in arcane/older Photoshop versions...

I haven't read about that in this topic. Am I missing some comment? Anyway, I have worked with photoshop all my life and never encountered this issue.

[debraspicher]

I agree, it is an issue that needs fixing. At 100%, it's not going to look too bad on my display because I have too much pixel density for it to be terribly blurry, but I do see that it's certainly not the raster crispiness we have come to treasure.

Hopefully @Chris B can advise.

[SrPx]

1 hour ago, javierr said:

Hi, @SrPx. Yes, that is the problem. Affinity Photo (in Photo Persona) it is not showing correctly a resized image layer. You can try the steps for yourself as described in my original question (first post).

 

I don't think Affinity is making a blur optimization because of low hardware or software, mainly because I have Photoshop installed in this same laptop and works correctly. I think there is a bug (which is strange so nooobody has pointed it in version 1 or 2) or Affinity is set to strange image screen perfomance and there is no place to change it.

 

In 100% zoom there is the same problem. Please try for your own with the steps described in my first post. In 100% you will see the image very small in screen, but still you can notice it is not sharp. Anyway, you can export it and check and zoom the exported file: it will be as blurry as you see in Affinity screen (unless you export it with lanczos, but that is a workaround, not a real solution).

 

I haven't read about that in this topic. Am I missing some comment? Anyway, I have worked with photoshop all my life and never encountered this issue.

It was not coming from any previous comment. I just remembered how with the  transform tool it would display some blur or bad edges in  PS (in an arcane version). But I have been using PS since 2.0 in '95. Don't mind that comment, it's irrelevant.

[SrPx]

I  made a test... not the one you mentioned (although I did the one you explained, before). Not sure is if it is relevant in any way, but this thread got me worried about certain (different) workflow. I resized to 10% (with 'scale layer' and with scale document, as not all apps have the same options) using lanczos 3 (except ClipStudio, I don't know what algo "high precision" refers to) with several applications (Krita, CSP, irfanview, PaintStorm, and the actual Photo). Disregard the color variations, I've just needed to reinstall all recently and some are configured yet as sRGB instead of all Adobe RGB color profile. But just copy pasting the mere pixels, no export involved in any case, and compared them with a linked-layer in APhoto, and another layer made by opening apart the pixabay full resolution photo, reducing it  there to 10%, select all, copy, and paste the pixels (making a pixel layer) in the new document. Well, the sharper image was this last one, as, it does seem to me that when the clipboard is from inside Aphoto (from another tab), if you move the layer, the pixels stay the same, no blur. All of the other sources, once I touch them, a slight blur occur, so, only at the end of the test I have realized this is not a fair test for the other applications. But thought I'd share, in case gives some extra detail. Or something. Layered psd attached, saved from AP (Photo beta).

So, I guess if one just paste the pixels, and immediately rasterizes the layer before even touching it with the cursor, this slight blur (which can be another issue, not saying it's related to  the problem you mention, but I thought I'd share), then perhaps the slight blur would not happen, or that's what's happening here. I don't know though, if  you have tested to apply an unsharp mask to the linked layer and see what happens.  And I am yet to test what happens by setting in preferences/general "prefer metafile when pasting from external application", I wonder if that helps the situation.

Edit: Yep... Even rasterized (with or without  the metafile thing, is not relevant) it happens when you move the layer the first time. Like it makes pixels to stay in intermediate positions. As a result, produces a slight blur in the image. If it ended in a position closer to the full pixel, then it looks a bit sharper. But if I set "force pixels aligning" (besides the snap icons, in the top bar) , then it does it right, no blur, pixel perfect. At least in this matter, that surely has not much to do with your problem. 

Edit 2: Now I just imported a file from Photoline, an exported psd. It was kind of sharp (from  the same woman file, lanczos 3 in that app). Definitely, it happens that just opening it, if I drag it without the force alignment pixels on, then-> blur (slight, yet noticeable). If I set that on (before touching the image after paste), no problem at all. Now, what has me even more intrigued is that if I open the pixabay image in another tab, ctrl + a, ctrl + v in the  testing document (127px wide document), scale the new layer to 10% right there, now, with force pixel alignment OFF, it does not produce the blur. While it seems to do it with copied pixels from other apps. As if the internal clipboard was protected from that to happen, or something. Anyway... I think I am going to keep that thingy "on", from now on (as solves the issue in my workflow, which is a lot simpler than yours: I am just a painter).

Edit 3: In some cases, applying an unsharp mask with the right  usual good values, many of the blurry images do look great (but yeah, not ideal).

 

some_test.afphoto

[javierr]

21 hours ago, SrPx said:

I  made a test...

Hi! I opened your file and all layers seems somewhat blurry... Not "photoshop quality" at all. Best one is lanczos for me, but being a copy-paste, all the big information is lost, so not use for me.

 

21 hours ago, SrPx said:

Edit: Yep... Even rasterized (with or without  the metafile thing, is not relevant) it happens when you move the layer the first time. Like it makes pixels to stay in intermediate positions. As a result, produces a slight blur in the image. If it ended in a position closer to the full pixel, then it looks a bit sharper. But if I set "force pixels aligning" (besides the snap icons, in the top bar) , then it does it right, no blur, pixel perfect. At least in this matter, that surely has not much to do with your problem. 

I am not sure if I understand you, but forcing aligning didn't gave me better results. Copying and pasting (from a new tab in Affinity to the test file, didn't make any difference. I can not copy from Photoshop and paste in Affinity.

I have tried this:

• Changing, in windows>Display>Scale and layout to 100% (I was using 150%). But no different results
• Export the Affinity Photo file to a PSD:
  • Preset: PSD (preserve accuracy) and Resamble: bilinear--> Same bad quality, but the Smart Object is preserved to its original size, and its fine
  • Preset: PSD (preserve accuracy) and Resamble: bilinear--> Good quality, better than Photoshop, but there is no Smart Object, just a layer (can not be resized bigger or will be blurry)
• Import the PSD file to Affinity Photo (just drag and drop): blurry, but the original image size is preserved, although not linked.

 

I have notice something, and just though could be a hint:

1. I usually have activated the option Perfomance>Tools>"Use mouse wheel to zoom". Ok, I have noticed that, when you zoom out (from about 170% to smaller percentajes) the image looks good for a milisecond, until the program calculates the new zoomed-out aspect. You can test with no mouse wheel also. I can play with it and all the times is doing that, so the problem is crearly in some processing of Affinity Photo.
2. If, with 300% zoom, you slightlly move the layer (move tool "V") you can see how the pixels "dance" trying to realocate the best position. This, in Phososhop, is unoticeable.

[SrPx]

That's their engine optimization thing, I believe.
What I meant (and testing now with a drag and dropped psd from a folder (exported from Photo from  the original jpg, as want to avoid using jpg format)) is that with "force pixel alignment" (I don't have the apps in English, sorry, don't know the exact term the UI uses) up above in the toolbar, all the time on, pressed, then that "pixels dancing" you refer to (which I see without this on) does not happen, not only moving the layer, but neither "dances" using the mouse wheel zoom, it moves then in steps of one full pixel, and at least here in my side, I don't see "blurrying" anymore (not zooming, not moving the layer), as I was seeing yesterday. But this is during operations, working with it. What I don't know (I'm 50, might have too tired eyesight (too much screen work in decades...), altho I believe my glasses are more or less ok) is if also internally during import, or using "transform" (as this is what  you are using), we are loosing more detail than if doing it with Photoshop. 

[NotMyFault]

3 hours ago, javierr said:

I usually have activated the option Perfomance>Tools>"Use mouse wheel to zoom". Ok, I have noticed that, when you zoom out (from about 170% to smaller percentajes) the image looks good for a milisecond, until the program calculates the new zoomed-out aspect. You can test with no mouse wheel also. I can play with it and all the times is doing that, so the problem is crearly in some processing of Affinity Photo.

A few post prior you said is has nothing to do with zoom level?

This observation is again something explained multiple times in the past: When you zoom, especially with View Quality auto, Affinity renders the image first in 1/2 resolution (or even worse). This sometimes leads to the false impression of sharper images. After some time (this can be split seconds to multiple seconds), Affinity renders in full resolution, often producing the impression of a blurry image. I once made a screen recording, and exported every single frame (50 per second) to see when exactly the transition happened.

 

 

Edited December 20, 2022 by NotMyFault
Added link to old post

[NotMyFault]

3 hours ago, javierr said:

If, with 300% zoom, you slightlly move the layer (move tool "V") you can see how the pixels "dance" trying to realocate the best position. This, in Phososhop, is unoticeable.

this is another big fundamental difference between Affinity Photo and all other Apps. Affinity considers the exact (fractional) position of all layers, and does not enforce any kind of pixel alignment. So moving a layer by 1/10th of an pixel width gives different renderings every time, If you have structures in the image with a resonating frequency (contrast lines with about 0.1-2.0 px distance), this can make a dramatic change in rendering.

Affinity does not actively relocate to best position. The user can do. Photoshop probably renders images differently. 

 

The linked post shows a test file explaining the effect:

Here is a test file, showing a checkered rectangle of 256x16px, copied 9 times and shifted by 0,1px in x-axis.

 

View at 800% (ctrl-4 / cmd-4)

With VQ Bilinear:

The same file with VQ "Nearest Neighbor":

This looks sharp, despite the image is actually blurry by resampling. Starting with 0,6px misalignment, you see that the squares are shifted by 1px to the right, so you swap blurriness by a different rendering issue. It is physically impossible to get a sharp image if it's misaligned (using pixel based display technology with fixed DPI).

Exporting these images with matching resample-settings will create the exact same image visible in Photo.

[javierr]

14 hours ago, SrPx said:

all the time on, pressed, then that "pixels dancing" you refer to (which I see without this on) does not happen

Yes, you are right. They don't dance. But the zoom in-out test is the same for me: it gets crispy just for a sec, and the the blurryness again.

 

14 hours ago, NotMyFault said:

A few post prior you said is has nothing to do with zoom level?

I mean that, once the program finished its calculations of the new zoom, the image gets blurry again. The crispy look only last a blink. But, I also have noticed that if you set the zoom to 192%, the image is perfectly crispy! Even better than in Photoshop at 192% zoom. It has some sense, as the original image is 1920px wide, then reduced to 10% so is 192px wide now. A zoom to 192% shows the image scaled perfectly, but it is not use.

 

14 hours ago, NotMyFault said:

This observation is again something explained multiple times in the past: When you zoom, especially with View Quality auto, Affinity renders the image first in 1/2 resolution (or even worse).

Thank you for the explanation, you are much nore technical than me but, what is this for? I mean, If the problem is not my hardware (and I guess is not cause more people seem to have the same problem), then Affinity has a problem (at least, I think having blurry images IS a problem). I don't care what the algorithm of Affinity is calculating: the result is bad. I know it is bad cause Photoshop gaves me better look and export. If I can not get a clean image, I can not work with it: I can not paint on it, I can not export it. What really really strange me is that nobody has commented it before. And don't get me wrong: I would love to work with Affinity software and forget Photoshop.

 

14 hours ago, NotMyFault said:

this is another big fundamental difference between Affinity Photo and all other Apps

The dancing pixels is not a problem: I understand the software is calculating something. What worries me would be the final result when you stop moving the layer. I know that Affinity works different as other apps, ok, I don't have problem with that, I am not a developer and not a technician, so I don't understand, don't have to care; I just want a good result.

 

Opening Affinity Photo today, I see there is a new update (but strange, cause it says update "from 2.0.3 to 2.0.3"...I guess it is a smaller fraction). I have tested it again and the problem persists.

[NotMyFault]

3 hours ago, javierr said:

Opening Affinity Photo today, I see there is a new update (but strange, cause it says update "from 2.0.3 to 2.0.3"...I guess it is a smaller fraction). I have tested it again and the problem persists.

Well, what you describe as bug is mainly the consequence of a fundamental design decision from the developers of Affinity Photo. I would be very surprised if Affinity decides to switch positions. All I can do is explain the reasons for differences, and what you could do to improve export quality. Affinity stated more than once that consequences of design decisions are not accepted as bug reports. 

Affinity is not simple clone of Photoshop. It is an individually developed app in the same functional domain, with a certain level of import / export compatibility to Adobe apps. As PS did set the standard for imaging software about 40 years ago, some similarities are unavoidable. Never the less, at some point you need to accept the differences, or file feature requests to close the remaining gap. Calling every difference a bug does not help.

 

 

[SrPx]

What worries me (as a finding on my own, surely already addressed here) is that yesterday I was doing a test comparing only internally two matters (blurring and the below issue with reduction quality)... I mean, not only PS, most apps I use that do image editing use the best algos for scaling/transforming a layer or a pixels selection. Yesterday I reduced that woman image to a very tiny resolution (127 px and 67px) and realized that the result was significantly poor by using "transform" panel or just resizing with the move tool handlers, compared to doing a document resize, in lanczos 3. As I say, in other apps there's no difference; it seems they use their top algorithm for layers and selections scaling up or down, too. But comparing pixel by pixel , the face contour and other features were definitely worse (not by a little), quite worse edge quality, than the document resized with Lanczos 3  to the same pixels size.  This means that at pixel level, we have a significant decrease of quality, and not sure if  only for tiny pixel sizes. Anyone knows a way to avoid this ? I was still using beta 2.0.3.

Probably it is only happening noticeably when reducing a relatively large image to a tiny 70x100 px image in a single go. And maybe would not happen in more normal size reductions. As anyway, I use Aseprite for pixel art-sprite animation, or can use any app or utility for extreme low res reductions like this. The fear is if it's decreasing all image sizes quality in general, in reductions/scaling towards not so tiny sizes.

Edit: Javierr: I had not noticed simply because I am usually working with very high res files, lately (and usually). 

[javierr]

5 hours ago, NotMyFault said:

Well, what you describe as bug is mainly the consequence of a fundamental design decision from the developers of Affinity Photo.

Hi, after trying other programs (Krita and Gimp) I am starting to understand what is happening here (or that is what I think) and it is a 180 degrees change view... Please correct me if I am wrong:

As you can see in the attached image, Krita, Gimp and Affinity Photo show a slightly blurry image. So it is not only Affinity Photo. That makes me think that Photoshop makes some improving (sharping?) on the resized layer. Could it that be? So Photoshop is the "tricky" one? Maybe I am so used to Photoshop that I think that was the proper behaviour? Could be something like that? In that case, we need Affinity developers to fix the Export Persona to show the same as Photo Persona (and which is it already fixed in macOS, as @Chris B said).

By the way, I was unable to get Gimp show a 300% zoom like the rest of programs. "View" is set to "Dot to dot" (with it off, it is even larger) and in Preferences>Interface>Display I have even calibrated the resolution with a ruler, but haven't got it "fixed" (if there is something to fix).

 

21 minutes ago, SrPx said:

Yesterday I reduced that woman image to a very tiny resolution (127 px and 67px) and realized that the result was significantly poor by using "transform" panel or just resizing with the move tool handlers, compared to doing a document resize, in lanczos 3

Could you paste the results here? Does the resized document in lanczos 3 keep the whole information on the woman layer? (I mean, if could it be made bigger again).

[NotMyFault]

 Using the transform panel and rasterizing will always use Bilinear resampling, which is the “softest” of all.

If you want to get control over which resample method is used, you must use Document->Resize or Pixel Art Resize (for upscale only)

[NotMyFault]

49 minutes ago, javierr said:

That makes me think that Photoshop makes some improving (sharping?) on the resized layer. Could it that be?

Absolutely. 
If you want to to sharpen / improve contrast, you can do this in Photo, but you need to do it manually by adding adjustments and/or filters.

[SrPx]

In my way of painting, working with linked layers makes less sense, it's more about rasterized pixel layers.  But I totally get that the smart way to work with them for anything else is linked files or non rasterized layers.

I tried that (the  transform with rasterized layers scaling down a lot) yesterday. Reducing to such tiny scale I'm sure that in  the case of other apps, scale/transform uses lanczos or bicubic sharper (yeah, maybe all they do internally is apply a sharpen). But in  this test, doing a careful unsharp mask and some other slight filter to help does not end with such a good image as a simple full document resize using lanczos to  that exact same tiny resolution. I'll do many more tests, mostly at higher resolutions, which are more realistic use cases for illustrators (pixel art uses other techniques, anyways). But I won't bore people with them here, as that would be going too off topic here, it gets out of the scope of the topic, I suppose. 

[javierr]

18 hours ago, NotMyFault said:

If you want to get control over which resample method is used, you must use Document->Resize or Pixel Art Resize (for upscale only)

Fine, although that would break a linked layer workflow, so... (see below)

18 hours ago, NotMyFault said:

Absolutely. 
If you want to to sharpen / improve contrast, you can do this in Photo, but you need to do it manually by adding adjustments and/or filters.

Ok. Understood. I attach some testing: As I use linked layers, I would use "Live Filters", which are non destructive; using "Unsharp mask" I have managed to replicate the same look as Photoshop has by default when resizing. I leave in the screenshot the values used in Unsharp mask" just if you want to try.

So ok, Affinity has a different way to get the same result, although it takes more steps, but, on exchange, you can be confident that the layer is not corrected by any automatic improvement: you do it if you want it, with adjustments or filters. I think I prefer this clean and non tricky approach when working.

Thanks for the help!

[SrPx]

On 12/21/2022 at 9:11 PM, javierr said:

Could you paste the results here? Does the resized document in lanczos 3 keep the whole information on the woman layer? (I mean, if could it be made bigger again).

To keep the image able to be up-scaled (as really, it is simply handling a high res image all the time) I guess depends on you not flattening stuff and on the file format chosen. If it is supported by the format, then why not: the sharpening filter or algorithm for scaling would not destroy that. The way of working I am interested in making sure is flawless is a bit different (I am an illustrator/painter, comic artist, also do graphic design and game art, 2D for 3D, etc. But not a photographer and rarely work with Publisher). Not about linked files or layers that keep the original size like PS's 'smart objects' that you can up-scale. But with pure pixels, 1:1. To work more at raw pixel level , in the way several painters do, but also 2D game artists (in high and low resolution).

I've made some tests, and although I might have got it wrong, my conclusions about  this aspect in Photo are more optimistic than I expected when I found the issue.

I have hidden the whole comment below, as rarely would be of interest for people other than the very curious about the matter, and it is just the experiments of someone not versed in Photo, but mostly in other apps. They're my findings, though.

TLDR/conclusion, anyway: Using 'force pixel alignment' and 'move full pixels' is the way to go (imo) for pixel art work, or other game art, texture work, or UI pixel perfect stuff that needs to be raster for some reason...and etc, working at pixel level needing a very 1:1 visual output as you work. This and working very much zoomed-in.  The export of both a high res or low res file (always speaking of working directly with pixels, not linked files or non-raster layers) as a PSD with Lanczos 3 Non-Separable always makes every layer look equally (very) sharp (but for the ringing, not good for pixel art). But I finally am going to opt for Lanczos "Separable" (which in theory is also faster in performance in export) as besides it produces less or no ringing, it gives more flexibility, if you work applying live filters for sharpening (ideally unsharp mask). Photo seems to keep more what you are setting in those filters (not even needing to flatten them previous to export) by using separable in the psd export. But I guess the non-separable option (which might be too sharp for some purposes) gives a nice sharp result without caring for any filter. And that's good for many, many users that won't know or won't care, and myself for other fast projects. 

Yet though, I hope some day we'll have a way to set which algorithm to use (as a setting to set and forget) for the raster layers manual resize, and "transform" panel, so to work with a  more 1:1 pixel output (or as close as it could get), or more adjustable quality as we edit or paint things. But till then, I can see the workflow I mentioned perfectly viable, it's no biggie. A bunch of Photoshop users will notice what we noticed in their first contact with  the app, so I hope the conclusions in this thread will give them some peace of mind.

Spoiler

Yes, you avoid the "pixel dance" by setting "force pixel alignment", if that bothers you, this leads to a bit lower positioning accuracy (it gets more "jumpy", as you move, but that's not a problem, it's very slight even in this super low res, pixel art resolution), while moving the layer. If also set "move full pixels", then the layer movement is even more "by steps" (even less smooth positioning), but it has the big benefit of not generating semi transparent pixels (terrible in pure old school pixel art) in the edges of  the layer, once the  movements stop. Whether if this is also translated to the final export, it is something I have not yet tested. Photoshop had also an "optimized" view, for performance (it's been a while since I last used PS) for the transform tool. But once you committed the transform (enter or double click), you would see the final quality (as exported) while editing, which already is 1:1 with the final export. It's a different technique (I kind of prefer the "wysiwyg" or "1:1" approach; but I know this is optimization, and it's different).

In the exported version there are no issues. We are not losing quality in any way, the "cooked" result is good. If you export a PSD  in lanczos 3 non-separable in the export, -and keep in mind this is a test where I have reduced that high quality big  image of a woman's face, to a tiny 67 px wide layer. It's an extreme situation, to see the hardest case- it all gets sharp and nice, maybe a bit too sharp even for something with artistic hard edges. But definitely sharp. Kind of (unless I did something wrong)  it seems to me that the clever way to export is to PSD Lanczos 3, but "Separable". It is the fastest algorithm of the two (I read it in other sources), but I don't say it for this, as Photo exports super fast anyway. It is because, at least here on my end, I am getting much more flexibility in such export. As mentioned earlier, but I just verified it for my type of output, the blurry thing is sort of solved adding a live Unsharp mask filter with the right values (usually 0.5 radius (and up to 1px if needing more, for some reason), anything close or past 1,0 as factor, and 0 as threshold, though, again, depends on the image, its size, and the effect you want).

This with the live filter, once exported gets a really similar look to a "document resize" using Lanczos3 non-separable, while exporting it with Lanczos 3 Separable. Having set the PSD export setting as Lanczos 3 Separable, seems to allow more flexibility. As export as non-separable just right away made all samples (layers) very sharp (in this low res, a bit too sharp, some ringing appearing), the export with Separable showed some smooth, some sharper, to the different degrees I had set with different Unsharp Mask live filters on each raster layer. Not even needing to set "rasterize all" in the export settings. Indeed, the best workflow (for this specific purpose!) is to neither rasterize the layers while editing in photo just before export. Just keep these live unsharp mask filters. As then whether you use in the PSD's export settings (for everything) "keep precision" or  "keep editing capability" (it does work with both the same), and leaving the live filters as they are, it exports all the sharp levels you wanted, respects  those live filters and their values. Resulting the sharpest just as an image resized with lanczos3 non-separable, as if that layer alone had been exported with non-separable. But with separable, keeping so more the subtlety that you were applying with different live filters. I don't know why is this, or if it is user (my error) error, but no matter how I try, the results I am getting are these.  Of course, some live filters and layer FX need flattening/rasterizing for certain formats and purposes. I'm only speaking about this specific case.

So, I'm concluding that for someone wanting  this level of control over the pixel output (and not dealing much with linked layers, or keeping the layers capability to be up-scaled again, etc, as I usually don't need it) one can work with "force pixel alignment" and "move keeping full pixels" for pure pixel art work, or only with "force.." if not enjoying the pixel dance, or retouching a game texture that must be very 1:1 at pixel level, and generally speaking, seems can use the live sharpening filters, at least for what is scaling down layers. But using (IMO) PSD's Lanczos 3 Separable.  I insist on PSDs because it is a format every client or etc individual doing a big project tends to prefer (a lot of people will ask for a  tiff or PDF, not having PS or sth that can open PSDs).  And because sadly it is the one way to do i/o with raster files with layers. Tiff can support layers somehow, but many apps will just flatten it. We can only export it flattened, but in my tests it indeed respects all the sharpening-smoothing levels of the live unsharp masks I mentioned before, with lanczos 3 Separable.

Yes, PS has a setting (pasted screenshot by Javier much above) to set how you do want the layers (and etc) resize, with which algo (bicubic sharper, nearest neighbor, etc (I think it has no lanczos, yet  )). Which is kind of preferable if you want to see a final result 1:1 while you work, even  if -in PS- you know that while doing the transform, or being the transform box active, what you view is a degraded optimization, but the rest of the time, you are seeing it just how it will look as an exported file (this is great for painting).

But...that said, using some live unsharp mask, you are getting almost 1:1 with the exported. I suppose the very slight difference even with that filter is way more evident (more than a usual scenario) when you are reducing in one step a 1900x1500 px layer of a detailed human face to a tiny icon/thumbnail/texture of 67 pixels wide. Meaning, with the usual layer resizes we probably won't be able to tell.

From what I tested, what we see is not all the story, the real internal data has more quality (yep, we are seeing an optimization, as said many times), will look great once exported.

And yes, all my tests have been with maximum rendering and quality settings in preferences (no auto settings either).

So, all good. Sorry if I have repeated stuff you both already said, or if I am being dense/dumb/ignorant.  I haven't used Photo as my tool for "everything" raster until recently. I have been years using it as a companion tool for final export and file preparation (for industry standards, etc), while using Clip Studio, Krita and other apps as the main work horse. 

As this is a tad long, maybe only of interest to you two, or one of you, or none ( ) , I am setting this hidden by the spoiler tag. So to let ppl know is unimportant, will leave just a TLDR paragraph out.

 

[NotMyFault]

On 12/21/2022 at 9:11 PM, javierr said:

Could you paste the results here? Does the resized document in lanczos 3 keep the whole information on the woman layer? (I mean, if could it be made bigger again).

That is impossible by the laws of physics:

• if you reduce the size of a document by e.g. to the half and rasterize using one of the offered resample methods, you will loose the „higher frequency“ information (think of high-pass filter). It cannot be „recovered“, neither by sharpening filter or any other filter. 
   
• you can scale-up the image again, but it will not be the same as before.
• Sharpening can boost edge contrast, giving the impression of of a sharper looking image, but never brings back lost details.

In theory it could be possible to create a resample method which preserves higher frequencies and drop lower frequencies, but this would create other loss of information (e.g. colors of smooth gradients).