Blending refactor and improvements #1065
Description
A meta-issue to track the refactoring and further improvements to the blending.
Main work to get to a new acceptable state
- Initial refactoring: Simplify blending #1002
- Post-processing API: Add support for post-processing effects #1114
- Implement FXAA: Add support for FXAA and DDAA #1117
- Implement new blending mechanics and API: Implement material.alpha_mode to replace .blending and .transparent #1144
- Set
material.aaoff by default, avoiding a range of artifacts: Turn off material.aa by default #1171 - Use a proper output-pass using cubic Mitchell kernel. Improve the interpolation filter in flush #1121
- Use a default pixel_ratio that's the same as the screen, or only 1.5 on regular screens. Add renderer.pixel_scale #1174
- Docs on dealing with transparency: Add docs for blending #1124
- Docs for aa: Add docs on aa #1173
- Review docs, add tips on how to deal with certain aritfacts (see The advantage of the
ordered2blend mode and how to go without it #1003). Also check docstrings ofaaprops. - Fix
.snapshot()so it uses the correct texture (see Add support for post-processing effects #1114 (comment)) Fix that snapshot may not include the last effect #1126 - Enable weighted blending between calls to render #1148
- Release.
Further enhancements
- Use half-float for internal render textures to avoid loss of precision (all ops are in linear space, but data is stored in srgb). Use rgba16float for intermediate color textures #1125
- Transmissive transparency Physically based transparency #974
- Dither can feel less noisy if the noise is static, which we can get with an object-id that is an index in the scenegraph, plus some element index. See ST paper. -> looked into this but we go with good blue noise anchored to screen.
- Revisit Changing
get_render_infoupon camera updates #1078 (comment) - Separating ”Pick_Info“ Generation from the Main Rendering Pipeline #748
- Allow reading depth values Access to depth buffer #320