Skip to content

Conversation

almarklein
Copy link
Member

Ref #1065

As planned, turning off shader-based aa.

We can afford it, because we have pretty good post-processing anti-aliasing now.

It avoids a range of artifacts by avoiding objects that produce a mix of opaque and transparent fragments.

@almarklein almarklein requested a review from Korijn as a code owner August 23, 2025 21:28
Korijn
Korijn previously approved these changes Aug 24, 2025
@almarklein almarklein changed the title Turn of material.aa by default Turn off material.aa by default Aug 24, 2025
@almarklein
Copy link
Member Author

I did some testing on various monitors today, trying different combinations of pixel_ratio, ppaa, and material.aa.

My proposal is:

  • Keep the behavior where we use SSAA on non-hiDPI displays
    • Well, actually fix/improve, see Rename alpha-method 'composite' to 'blended' #1172
    • It gives a good baseline for preventing aliasing for any object.
    • It seems 'fair' / consistent; the internal texture size is more or less the same, regardless of the monitor being used.
    • People can set renderer.pixel_scale=1 if they want to squeeze performance.
  • Enable ppaa by default.
    • It's cheap and helps a lot to get a smooth result.
  • Disable material.aa by default.
    • It does not improve the result much if the above is used.
    • Except perhaps text, where slight variations in the perceived width of the stems of am H can stand out.
    • I'm still on the fence between enabling aa in text by default, or simply enabling it in all examples and docs.
    • Though tools that work in 2D like fastplotlib should probably enable aa on objects, because it's cheap.

@almarklein almarklein merged commit 5e79f29 into main Aug 26, 2025
13 checks passed
@almarklein almarklein deleted the aa-false branch August 26, 2025 10:20
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants