potential bug with fill layer alpha blending

[NotMyFault]

Hi,

i created a test file to compare how alpha blends in different situations, especially "color opacity vs layer opacity".

The file

• contains 16 tiles
• 4 "large" tiles showing
  • top: vector shapes,
  • bottom: fill layers,
  • left: raw,
  • right: copy of left rasterized & tim
• 4 "small" tiles, showing left:
  • red left color opacity   / layer opacity : 50 / 100,
  • right 100 / 50
  • blue top 50 / 100,
  • bottom 100 / 50

As visible, the fill layer blend result differs from vector blend result, and rasterized layers blend results (bottom left corner)

I strongly assume a bug with fill layers alpha blending - except someone from Affinity can provide a good explanation in layman's terms (or exact math formula)

 

color vs layer opacity vector vs fill bug.afphoto

Edited August 25, 2021 by NotMyFault

[NotMyFault]

To give this another direction: It would be really really great if Affinity (James?) would provide some tutorials how all this combines:

• Color opacity
• Layer opacity
• Fill opacity (from Layer FX)
• Blend differences wrt alpha /opacity of Fill layers vs. all other types of layers

And other areas or arcane secrets like

• Brush Opacity
• Brush Flow
• Brush Accumulation
• Fun explosion: In combination with all these alpha blend options mentioned above

And more arcane secrets of layer blending

• layers following normal alpha compositing blend rules (pixel, curves, ...)
• layers able to overwrite blend rules (PT Filter, alpha-impacting adjustments levels , curves, channel mixer: just set/define new alpha value)

How all this integrates

• layers flat atop each other
• layers nested to description of ordinary layers
• layers nested to thumbnail of ordinary layers (masking / clipping position)
• layers nested to description of group layers
• layers nested to thumbnail of group layers
• layers nested to stacking groups (alpha reduced to 0 / 1)

Edited August 25, 2021 by NotMyFault

[Gabe]

Hi @NotMyFault,

I will get this logged and see if it's right or not, but I'm afraid we will not be providing any formulas for how this is calculated. 

[NotMyFault]

Maybe i found a probable cause:

When reducing layer opacity of fill layer to 50%, it becomes much lighter vs. all other methods.

You can spot this by disabling all red layers.

maybe related to (missing / double) gamma correction in blend formula.

0.7^2.2 roughly equals 0.5

if i change opacity to .7 it gives expected color

[NotMyFault]

Only a Fill Layer opacity will transform the alpha channel with a gamma of 2.0 which seems to be wrong.

As far as i know, alpha channels does not apply gamma correction, see comments by Gabe

and:

https://en.wikipedia.org/wiki/Alpha_compositing

Note that only the color components undergo gamma-correction; the alpha channel is always linear.

 

[IPv6]

1 hour ago, NotMyFault said:

As far as i know, alpha channels does not apply gamma correction, see comments by Gabe

Still, they are applied - in RGB32 mode // its an unfortunate decision made by Serif 

[NotMyFault]

2 hours ago, IPv6 said:

Still, they are applied - in RGB32 mode // its an unfortunate decision made by Serif 

Maybe we are talking about completely different topics. I strongly assume Affinity don't use Gamma correction on Alpha channel for layer blending in any color mode - except the (faulty) case of fill layers. Specifically what is explained in the wikipedia link about Alpha Composting.

I did not relate to other situations, e.g. converting between 8/16 and 32 bit modes, or channels panels / channels mixer etc. I've read your post in the linked thread, but couldn't follow your argumentation. I will try to make some artificial test documents on my own, this helps me to understand.