View Quality: Bilinear (Best Quality) setting shows the images blurred if X, Y are non integer values

[carl123]

There was a thread discussing a similar issue yesterday and today but it seems to have disappeared.

Anyway...

If the X and Y px values of a pixel layer are integers then I see no difference switching between Bilinear and Nearest Neighbour in Edit > Settings > Performance (e.g. at 500% zoom)

But if either the X or Y values have decimals then Bilinear produces a blurrier image, whereas Nearest Neighbour displays exactly the same image as when using integer values

Does anyone know what may explain this as I would expect to get the best image quality when using Bilinear but all I can get is the same quality as Nearest Neighbour or a worse image quality (i.e. blurred)?

Note: I am only taking about viewing a pixel image on screen, not exporting or printing it
 

Windows 11

[Designer1]

I can agree with that. The question is, what exactly is the solution? When I open an image with typography, the image looks blurred. This only affects images and not vector graphics.

[mopperle]

I think thats why there are different options for interpolation.

Nearest neighbour works best with lineart and illustrations, while bilinear is meant for images. IMHO AfPhoto sees a screenshot of text/dialogboxes as lineart. Using nearest neighbour makes a screenshot much sharper.  An so it is when I open the attched screenshot in AFPhoto. Using A picture like the dog (attached) I see no difference.

In Adobe Photoshop you have a third option (with 3 variants): bicubic, which is set as default, but needs a longer processing time. But I see no difference at all between those three. I assume that this setting is used for export only.

Some sources regarding this:
https://www.hollandlitho.com/image_scaling_limitations.html

https://www.salierdruck.de/was-ist-interpolation (German only)

 

[Ldina]

1 hour ago, mopperle said:

Nearest neighbour works best with lineart and illustrations, while bilinear is meant for images. IMHO AfPhoto sees a screenshot of text/dialogboxes as lineart. Using nearest neighbour makes a screenshot much sharper.  An so it is when I open the attched screenshot in AFPhoto. Using A picture like the dog (attached) I see no difference.

I want to be sure I understand this properly.

Do the Performance Options in Settings/Preferences (i.e., View Quality, Dither Gradients, Use Precise Clipping) ONLY affect the display of data to the monitor? In other words, changing these options in Settings/Preferences does not alter the data inside the file at all?

If that's the case, then I can set them to whatever I want without affecting the file data. Those settings and how they are rendered on screen may affect my judgement when sharpening, creating gradients, etc, but the same, exact adjustments applied to a file will be identical regardless of these performance settings?

What about Exporting and Printing? I assume screen rendering is immaterial, and the data in the file will be exported or printed, regardless of Performance Settings?

When resizing a file, I recognize that the Resampling method WILL have an impact on the final pixel data.

Thanks.

[mopperle]

Easy to check: when I load e.g. my screenshot file above and switch between the interpolation options under Settings/Performance, I can close it without being asked to save it, means nothing in the file has changed. So the setting is just something influencing the rendering on the screen. If I want to change the interpolation for the file, I have to do this via the export function.

In Adobe Photoshop, it is a bit different. The settings there apply only to the export function. For rendering on the display, it seems that they use a "hardcoded" setting. Cause I do not see anyy difference on the screen when switching between the various options. But not 100% sure about this.

 

[Ldina]

42 minutes ago, mopperle said:

So the setting is just something influencing the rendering on the screen.

Thanks, @mopperle. That's what I was assuming, but I appreciate the confirmation. 

[NotMyFault]

My reply to that question got lost yesterday.

 

 

 

[NotMyFault]

7 hours ago, carl123 said:

Does anyone know what may explain this as I would expect to get the best image quality when using Bilinear but all I can get is the same quality as Nearest Neighbour or a worse image quality (i.e. blurred)?

You get what you have asked for from Affinity. The result may seem counter intuitive. Affinity brutally applies the (correct) mathematical formulas without any safety net.

Users expect from today’s apps to automatically adjust positions to whole integer, or even disallow positioning to fractional positions. Affinity lacks seat belts, airbag, anti-lock breaks, and all other modern assistants. The UI amplifies the issue.

In theory Affinity could detect those simple case of misalignment and move the layer. In case you add other complexities or use other methods (scale by 1 bit up or down, skew, rotate, …) it becomes impossible to predict if the user intends the effect or did a mistake.

 

[NotMyFault]

Another tutorial explaining different Resample options in situations like

• rendering inside app (zoom level)
• resizing layers inside app by move tool or rasterize (always bilinear, even if NN chosen in performance)
• resizing and Resample method at export
• resize and Resample document (all layers)

 

[Designer1]

13 hours ago, NotMyFault said:

My reply to that question got lost yesterday.

 

 

 

I read your comments with interest. In practice, changing to "Pixels 6" does not result in any improvement in display quality. In fact, this represents a change to Affinity’s default settings. Moreover, I do not work with pixels, but rather with millimeters—and with points when it comes to typography.

 

13 hours ago, NotMyFault said:

My reply to that question got lost yesterday.

 

 

 

I read your comments with interest. In practice, changing to "Pixels 6" does not result in any improvement in display quality. In fact, this represents a change to Affinity’s default settings. Moreover, I do not work with pixels, but rather with millimeters—and with points when it comes to typography.

[NotMyFault]

1 hour ago, Designer1 said:

In practice, changing to "Pixels 6" does not result in any improvement in display quality.

It does not impact rendering quality in any way directly.
 

This only allows to spot misalignment of layers in transform panel which can cause unwanted blurring of pixel layers (and semitransparent edges of vector layers). And it requires to use px as document units, or temporary switch units.

 

 

[carl123]

19 hours ago, Ldina said:

If that's the case, then I can set them to whatever I want without affecting the file data. Those settings and how they are rendered on screen may affect my judgement when sharpening, creating gradients, etc, but the same, exact adjustments applied to a file will be identical regardless of these performance settings?

What about Exporting and Printing? I assume screen rendering is immaterial, and the data in the file will be exported or printed, regardless of Performance Settings?

Actually, I think screen rendering may be more important as regards the exported file than first thought but not in the way you would initially suspect

When set to Nearest Neighbour (Fastest) you see no difference (on screen) if looking at an image which has integer or non-integer X/Y values 

When set to Bilinear (Best Quality) you will see a difference. With the non-integer version of the image looking blurrier on screen which matches the fact that it will also be blurrier if exported (Due to the non-integer X/Y values)

Hence when set to Bilinear (Best Quality) I think your screen is giving you the best representation of what the exported image will look like. Which explains why you see a blurrier image with non-integer X/Y images.

Bilinear (Best Quality) is effectively trying to show/warn you that the image will be exported blurred and as such tries to give you the best rendering (on screen) of what the exported file will ultimately look like

 

[NotMyFault]

16 minutes ago, carl123 said:

Bilinear (Best Quality) is effectively trying to show/warn you that the image will be exported blurred and as such tries to give you the best rendering (on screen) of what the exported file will ultimately look like

It is just the industry wide compromise used in almost every app how to deal with rendering of pixel content when zoomed.

Affinity always Resample both in app and at export (and when rasterizing, when using brush tools etc), it has no „shortcut“ to not resample for the extreme rare case of showing 100% zoom level, perfect pixel alignment, one fully opaque layer.

It is unavoidable and in no way specific to affinity.

every app supporting zooming, multiple layers of different DPI or rotation, layer blending, transparency, MUST use resampling to produce a rendering.

its the core meaning of rendering - transform the color values of content into a meaningful representation for the display, and calculate the required color values of each individual pixel (in display resolution).

Only if you restrict yourself to ms paint (in the first version), no layers, only 100% opacity and fix zoom level to 100% you don’t need resampling
 

[Designer1]

54 minutes ago, NotMyFault said:

It is just the industry wide compromise used in almost every app how to deal with rendering of pixel content when zoomed.

Affinity always Resample both in app and at export (and when rasterizing, when using brush tools etc), it has no „shortcut“ to not resample for the extreme rare case of showing 100% zoom level, perfect pixel alignment, one fully opaque layer.

It is unavoidable and in no way specific to affinity.

every app supporting zooming, multiple layers of different DPI or rotation, layer blending, transparency, MUST use resampling to produce a rendering.

its the core meaning of rendering - transform the color values of content into a meaningful representation for the display, and calculate the required color values of each individual pixel (in display resolution).

Only if you restrict yourself to ms paint (in the first version), no layers, only 100% opacity and fix zoom level to 100% you don’t need resampling
 

For me, all these theoretical explanations are irrelevant in practical work. In my opinion, Adobe Illustrator exports a much better, smoother, and sharper image in terms of typography compared to Affinity Designer. All the discussions about pixels and such are meaningless to me as a user because there is no setting that allows for exporting the same quality as Adobe Illustrator. On top of that, in Affinity Designer 2.6.2, the exported images appear blurry. How to fix this issue is still completely unclear to me.

[NotMyFault]

45 minutes ago, Designer1 said:

For me, all these theoretical explanations are irrelevant in practical work. In my opinion, Adobe Illustrator exports a much better, smoother, and sharper image in terms of typography compared to Affinity Designer. All the discussions about pixels and such are meaningless to me as a user because there is no setting that allows for exporting the same quality as Adobe Illustrator. On top of that, in Affinity Designer 2.6.2, the exported images appear blurry. How to fix this issue is still completely unclear to me.

We know your position from many older posts. It is ok. 

but if you want to stay away from these „theoretical explanations“, we can only give the practical advise to stay away from Affinity products and continue to use competing apps. The issues won’t be fixed in predictable time, if it will be fixed at all.

A portion of the „issue“ is in how you personally perceive the difference, and keep refusing to produce „tangible“ documents where others are able to reproduce what who perceive. It will be cross-talking forever. I see no common level of communication or understanding between your very personal expectations and the software / formal discussions what apps can do or not. Affinity is no 1:1 clone of Adobe. There are millions of minor differences in workflows and the results when it comes to those level of details which may be live or death for you but mostly irrelevant for others, or companies like Affinity who must decide on what features or bug fixes to prioritize on.
 

[Designer1]

2 hours ago, NotMyFault said:

We know your position from many older posts. It is ok. 

but if you want to stay away from these „theoretical explanations“, we can only give the practical advise to stay away from Affinity products and continue to use competing apps. The issues won’t be fixed in predictable time, if it will be fixed at all.

A portion of the „issue“ is in how you personally perceive the difference, and keep refusing to produce „tangible“ documents where others are able to reproduce what who perceive. It will be cross-talking forever. I see no common level of communication or understanding between your very personal expectations and the software / formal discussions what apps can do or not. Affinity is no 1:1 clone of Adobe. There are millions of minor differences in workflows and the results when it comes to those level of details which may be live or death for you but mostly irrelevant for others, or companies like Affinity who must decide on what features or bug fixes to prioritize on.
 

---

Just to be clear, I think Affinity Designer and Affinity Publisher are excellent apps. I really enjoy working with them. The only thing that bothers me is the export quality — as mentioned, it could be better. I should also mention that the price-performance ratio with Affinity is excellent. I own all three apps.

I think the export quality in Affinity Designer suffers from lower ppi.

PPI stands for Pixels Per Inch, and on smartphones, it refers to the pixel density – how many pixels are packed into one inch of the screen.

The higher the PPI, the sharper and clearer the screen looks, because the pixels are smaller and more tightly packed.

What applies to smartphones should also apply to the export of images.

[R C-R]

8 hours ago, NotMyFault said:

This only allows to spot misalignment of layers in transform panel which can cause unwanted blurring of pixel layers (and semitransparent edges of vector layers). And it requires to use px as document units, or temporary switch units.

Which makes me wonder why the default is not higher, at least to 2-3 decimal points.

[R C-R]

4 hours ago, Designer1 said:

PPI stands for Pixels Per Inch, and on smartphones, it refers to the pixel density – how many pixels are packed into one inch of the screen.

The higher the PPI, the sharper and clearer the screen looks, because the pixels are smaller and more tightly packed.

What applies to smartphones should also apply to the export of images.

PPI has meaning for physical things like computer or smartphone screens, but even on smartphones one can usually zoom in or out such that the pixels of the displayed image do not exactly match the pixels of the display, thus (as @NotMyFault mentioned above) resampling must occur to render it onscreen. IOW, the pixel density of the display itself cannot be changed, thus the necessary compromise.

Likewise, if a document that does not have 'pixel-perfect' alignment of all of its objects (IOW, every one of every object's pixels are not aligned exactly to the document's pixels) then the exported image cannot reproduce that without either anti-aliasing (thus leading to softening or blurring) or moving and/or resizing every misaligned object to eliminate the need for anti-aliasing.

Affinity does not do any moving/resizing so users must make sure everything is pixel perfect before exporting, which is why setting the display decimal point to a large enough value to see that is so often suggested.

[Designer1]

13 hours ago, R C-R said:

PPI has meaning for physical things like computer or smartphone screens, but even on smartphones one can usually zoom in or out such that the pixels of the displayed image do not exactly match the pixels of the display, thus (as @NotMyFault mentioned above) resampling must occur to render it onscreen. IOW, the pixel density of the display itself cannot be changed, thus the necessary compromise.

Likewise, if a document that does not have 'pixel-perfect' alignment of all of its objects (IOW, every one of every object's pixels are not aligned exactly to the document's pixels) then the exported image cannot reproduce that without either anti-aliasing (thus leading to softening or blurring) or moving and/or resizing every misaligned object to eliminate the need for anti-aliasing.

Affinity does not do any moving/resizing so users must make sure everything is pixel perfect before exporting, which is why setting the display decimal point to a large enough value to see that is so often suggested.

So pixel exact alignment of typography (X/Y) is not a problem. However, if the typography is in points, it cannot be scaled with pixel accuracy. Therefore, these settings (B/H), which are described here as optimal, cannot be realized in practice.

[NotMyFault]

The trick is to use 144 dpi or or integer multiples of 72. Then pt and px match. It was mentioned by another user in a different thread, pointing to an Affinity Spotlight article.

 

found it (from 2019 and then unfortunately forgotten. It must become a FAQ!)

https://affinityspotlight.com/article/hi-res-iconui-design-can-be-pixel-perfect/

 

[Designer1]

9 minutes ago, NotMyFault said:

The trick is to use 144 dpi or or integer multiples of 72. Then pt and px match. It was mentioned by another user in a different thread, pointing to an Affinity Spotlight article.

 

found it (from 2019 and then unfortunately forgotten. It must become a FAQ!)

https://affinityspotlight.com/article/hi-res-iconui-design-can-be-pixel-perfect/

 

 

Interesting. This is a very useful article. In conclusion, I would say that if you align everything pixel-perfect, then export it 3x as PNG and then reduce it to the required size - you get excellent quality. The only problem is that not everyone is familiar with this procedure. It also takes some time and work.