VSE: Text shadow blur / outline #121478

Aras Pranckevicius · 2024-05-06T14:35:17+02:00

2024-05-06 14:35:17 +02:00

Implements small part of #112698 and possibly other existing requests (from right-click select: https://blender.community/c/rightclickselect/dJfbbc/ and https://blender.community/c/rightclickselect/xTdbbc/):

Text got configurable shadow angle and offset (previously shadow location relative to text was fixed). Currently the shadow position is rounded to integer pixel locations (sub-pixel positioning might come later).

Text got optional shadow blur amount:
Shadow blur 0.1:
Shadow blur 0.4:
Shadow blur 0.8:

Text got optional outline option, with separate outline color and outline width. Outline is calculated using jump-flooding algorithm (similar to what compositor does for inpainting / double edge mask), which in my tests is faster than a blur based approach even at small outline widths. At some point it would allow even mapping a gradient or whatever across the outline.

Here's large font with 30px outline:

The same with 3px outline:

And a tiny (12pt) text with 2px outline:

TODO / Questions:

Test coverage.
Performance could be improved by doing all blurring / outline only in the area that the text actually occupies, not on the full image.
Found an issue in BLF where it accepts "how many color channels an image has" argument but proceeds to the the wrong thing afterwards. Not sure if should be fixed separately (or that functionality removed?)

Implements small part of #112698 and possibly other existing requests (from right-click select: https://blender.community/c/rightclickselect/dJfbbc/ and https://blender.community/c/rightclickselect/xTdbbc/): ![image](/attachments/32fcacd7-a299-433c-a84f-9a6a596de08a) Text got configurable **shadow angle and offset** (previously shadow location relative to text was fixed). Currently the shadow position is rounded to integer pixel locations (sub-pixel positioning might come later). Text got optional **shadow blur amount**: Shadow blur 0.1: ![image](/attachments/5ca3509a-97eb-424e-8969-231316278cd0) Shadow blur 0.4: ![image](/attachments/dcf6634f-710c-4730-8aad-867bf073cae2) Shadow blur 0.8: ![image](/attachments/1b371b09-fb78-4597-95aa-06d91ac4e450) Text got optional **outline option**, with separate outline color and outline width. Outline is calculated using jump-flooding algorithm (similar to what compositor does for inpainting / double edge mask), which in my tests is faster than a blur based approach even at small outline widths. At some point it would allow even mapping a gradient or whatever across the outline. Here's large font with 30px outline: ![image](/attachments/31e2905e-3c71-4055-9cfe-9a6d810252dd) The same with 3px outline: ![image](/attachments/c7dda39a-390e-45b7-b220-b9e98fc34c1b) And a tiny (12pt) text with 2px outline: ![image](/attachments/eb63aeb1-fae5-4cef-96a1-f56271a4b224) **TODO / Questions**: - Test coverage. - Performance could be improved by doing all blurring / outline only in the area that the text actually occupies, not on the full image. - Found an issue in BLF where it accepts "how many color channels an image has" argument but proceeds to the the wrong thing afterwards. Not sure if should be fixed separately (or that functionality removed?)

image.png

42 KiB

image.png

36 KiB

image.png

40 KiB

image.png

33 KiB

image.png

2.1 KiB

image.png

94 KiB

image.png

113 KiB

image.png

21 KiB

image.png

25 KiB

image.png

26 KiB

image.png

114 KiB

👍 3 ❤️ 11 🎉 2 🚀 2

Aras Pranckevicius added 3 commits 2024-05-06 14:35:27 +02:00

564d3318c0 VSE: Text strip gets configurable shadow angle and offset

df2e5f60a5 VSE: Text strip shadow can be blurred

6e044195b5 VSE: Text got outline option

hamza.SMA commented

2024-05-06 19:59:33 +02:00

First-time contributor

maybe or really i'm pushing too far , i'm dreaming having Shadows of the shadow meaning double outline ..(or more with custom color for each (color ramp ?))

Aras Pranckevicius added 1 commit 2024-05-07 10:23:34 +02:00

2a3f88d7d0 VSE: Text outline is now done via JFA algorithm

Advantages:
- "More round" outlines especially at large widths,
- Actually faster, e.g. 4K VSE resolution, outline width 30: outline
  calculation 110ms -> 40ms. At small outline widths, about the same
  performance as blur based approach.
- If someone wants to make outlines "fancy" someday (e.g. mapping a
  gradient across outline radius), this is trivially doable with JFA.

Aras Pranckevicius added 1 commit 2024-05-07 11:53:23 +02:00

d9c9c7d786 VSE: text shadow blur does not get too dark at edges when shadow color is bright enough

Gaussian blur code is not fully premultiplication-correct. So render
text as white, blur it, and then tint with shadow color when compositing.
Makes large blurs with small shadow opacity work better too.

Aras Pranckevicius added 2 commits 2024-05-07 13:19:16 +02:00

3e2b8b08e4 Merge branch 'main' into vse_text_effects

buildbot/vexp-code-patch-darwin-x86_64 Build done. Details

buildbot/vexp-code-patch-darwin-arm64 Build done. Details

buildbot/vexp-code-patch-windows-amd64 Build done. Details

buildbot/vexp-code-patch-lint Build done. Details

buildbot/vexp-code-patch-linux-x86_64 Build done. Details

buildbot/vexp-code-patch-coordinator Build done. Details

d199cfe108 VSE: initialize new variables of text strips when loading older files

And shadow offset/angle defaults that better match the previous hardcoded values

Aras Pranckevicius commented

2024-05-07 13:19:44 +02:00

@blender-bot build

Mikamiel commented

2024-05-07 15:23:16 +02:00

First-time contributor

Great feature! Just a remark: is it possible to smooth these jagged edges/artefacts and put some antialiasing?

Great feature! Just a remark: is it possible to smooth these jagged edges/artefacts and put some antialiasing? ![изображение](/attachments/56244604-69ef-4bfa-851a-babfb6443d6d)

изображение.png

88 KiB

Aras Pranckevicius commented

2024-05-07 19:02:17 +02:00

Just a remark: is it possible to smooth these jagged edges/artefacts and put some antialiasing?

@Mikamiel that "weird line" in the middle has nothing to do with shadow/outline (or even text), it's just that there was fractional scaling in the VSE preview window when I took the screenshots.

> Just a remark: is it possible to smooth these jagged edges/artefacts and put some antialiasing? @Mikamiel that "weird line" in the middle has nothing to do with shadow/outline (or even text), it's just that there was fractional scaling in the VSE preview window when I took the screenshots.

Aras Pranckevicius added 1 commit 2024-05-07 19:20:40 +02:00

df263e746e VSE: make outline width also be relative to font size, similar to shadow blur/offset

Aras Pranckevicius changed title from ~~WIP: VSE: Text shadow blur / outline~~ to VSE: Text shadow blur / outline

2024-05-07 19:28:38 +02:00

Aras Pranckevicius added this to the Video Sequencer project 2024-05-07 19:28:50 +02:00

Aras Pranckevicius requested review from Sergey Sharybin 2024-05-07 19:28:56 +02:00

Aras Pranckevicius requested review from Richard Antalik 2024-05-07 19:29:06 +02:00

Mikamiel commented

2024-05-07 19:32:42 +02:00

First-time contributor

That's what it was!)

Richard Antalik approved these changes 2024-05-07 21:20:32 +02:00

Richard Antalik left a comment

I was following the JFA, untill it got to AA part, which is bit blurry (HA!) to me. but I got the gist.
Technically both JFA and shadow can be optimized by only working on part of buffer, because you know the position and size of these elements.

Quite some time ago I wanted to add support for blending buffers of non-uniform size, as this would make using text strip much faster overall, but that would be bit more complicated.

I was following the JFA, untill it got to AA part, which is bit blurry (HA!) to me. but I got the gist. Technically both JFA and shadow can be optimized by only working on part of buffer, because you know the position and size of these elements. Quite some time ago I wanted to add support for blending buffers of non-uniform size, as this would make using text strip much faster overall, but that would be bit more complicated.

Sergey Sharybin reviewed 2024-05-08 10:07:17 +02:00

Sergey Sharybin left a comment

While testing I've noticed that the shadow is only moved by integer pixels, so if you animate shadow rotation it kind of jumps rather than having smooth anti-aliased transition. It is not something that necessarily needs to be supported by this initial PR, but it worth stating it as a known limitation in the PR description.

Found an issue in BLF where it accepts "how many color channels an image has" argument but proceeds to the the wrong thing afterwards. Not sure if should be fixed separately (or that functionality removed?)

I didn't look into details of what exactly is going wrong there, but it would be good if we the broken functionality got addressed separately. it'll make it easier to review.

While testing I've noticed that the shadow is only moved by integer pixels, so if you animate shadow rotation it kind of jumps rather than having smooth anti-aliased transition. It is not something that necessarily needs to be supported by this initial PR, but it worth stating it as a known limitation in the PR description. > Found an issue in BLF where it accepts "how many color channels an image has" argument but proceeds to the the wrong thing afterwards. Not sure if should be fixed separately (or that functionality removed?) I didn't look into details of what exactly is going wrong there, but it would be good if we the broken functionality got addressed separately. it'll make it easier to review.

source/blender/sequencer/intern/effects.cc

						
				@ -2706,0 +2801,4 @@

				    IMB_freeImBuf(tmp_out1);

				    IMB_freeImBuf(tmp_out2);

				    BLF_buffer(font,

				               out->float_buffer.data,

Sergey Sharybin commented

2024-05-08 10:01:30 +02:00

Can we pass nullptr instead of out->float_buffer.data?
To me it will be much more clear indication of what we expect to be happening here: we only expect the text to operate o na byte buffers, and we only write to byte buffer. So when we pass float buffer somewhere it reads as if it is expected that it might be a valid buffer.

Can we pass `nullptr` instead of `out->float_buffer.data`? To me it will be much more clear indication of what we expect to be happening here: we only expect the text to operate o na byte buffers, and we only write to byte buffer. So when we pass float buffer somewhere it reads as if it is expected that it might be a valid buffer.

aras_p marked this conversation as resolved

Aras Pranckevicius added 2 commits 2024-05-08 10:22:54 +02:00

dd7fa3b4a4 Merge branch 'main' into vse_text_effects

90933b8acd Cleanup: pass nullptr for float buffer to BLF_buffer to better indicate intent

Aras Pranckevicius commented

2024-05-08 10:25:52 +02:00

While testing I've noticed that the shadow is only moved by integer pixels, so if you animate shadow rotation it kind of jumps rather than having smooth anti-aliased transition. It is not something that necessarily needs to be supported by this initial PR, but it worth stating it as a known limitation in the PR description.

This seems to be general limitation of whole BLF system: the text effect code does pass floats to BLF_position, but then that one internally rounds them to nearest integer coordinate, and stores as integers for later rasterization. That is: all texts across whole blender are positioned at integer pixel coordinates, it seems.

I guess with some additional work we could rasterize the text for shadow at non-offset-at-all position, then blur it, and then manually place it where needed, with bilinear interpolation. But perhaps better done separately indeed (including perf optimizations that would do outline/blurring only on text area, and not the whole image).

> While testing I've noticed that the shadow is only moved by integer pixels, so if you animate shadow rotation it kind of jumps rather than having smooth anti-aliased transition. It is not something that necessarily needs to be supported by this initial PR, but it worth stating it as a known limitation in the PR description. This seems to be general limitation of whole BLF system: the text effect code does pass floats to `BLF_position`, but then that one internally rounds them to nearest integer coordinate, and stores as integers for later rasterization. That is: all texts across whole blender are positioned at integer pixel coordinates, it seems. I guess with some additional work we could rasterize the text for shadow at non-offset-at-all position, then blur it, and then manually place it where needed, with bilinear interpolation. But perhaps better done separately indeed (including perf optimizations that would do outline/blurring only on text area, and not the whole image).

Sergey Sharybin commented

2024-05-08 10:48:10 +02:00

That is: all texts across whole blender are positioned at integer pixel coordinates, it seems.

The BLF historically was done to draw UI texts indeed, where you do want to place it as pixel-perfect as possible. Is only much later when it started to be used for an effect type of a thing.

> That is: all texts across whole blender are positioned at integer pixel coordinates, it seems. The BLF historically was done to draw UI texts indeed, where you do want to place it as pixel-perfect as possible. Is only much later when it started to be used for an effect type of a thing.

Sergey Sharybin approved these changes 2024-05-08 10:54:02 +02:00

Sergey Sharybin left a comment

For the description I'd still suggest doing something like:

Text got configurable shadow angle and offset (previously shadow location relative to text was fixed). Currently it is rounded to integer pixels. It is possible to allow sub-pixel positioning for animation, but it is not planned for this first step.

For the description I'd still suggest doing something like: > Text got configurable shadow angle and offset (previously shadow location relative to text was fixed). Currently it is rounded to integer pixels. It is possible to allow sub-pixel positioning for animation, but it is not planned for this first step.

👍 1

Aras Pranckevicius merged commit d66598c16f into main

2024-05-08 11:13:33 +02:00

Aras Pranckevicius referenced this issue from a commit

2024-05-08 11:13:33 +02:00

VSE: Text shadow blur / outline

Aras Pranckevicius deleted branch vse_text_effects

2024-05-08 11:13:35 +02:00

Aras Pranckevicius referenced this pull request

2024-05-10 09:59:43 +02:00

VSE: Optimize text shadow blur / outline options #121644

Aras Pranckevicius referenced this issue from a commit