Page MenuHome

color type alpha channel not yet implemented (for attribute mix node)
Needs Information from Developers, NormalPublic

Description

Hello

It seems that the attribute mix node do not mix alpha's properly, for example, when mixing a color with alpha of 1 and color with alpha of 0, no interpolation is done, therefore the result is not accurate (same for every other caclulation method it seems)

Overall the 4th channel of this color vector type is straight out ignored during the calculation and i believe that this attribute mix node did not implemented the color type properly?

On a side note: there's missing RGBA to color and color to RGBA attribute separate node, therefore, for the time being, this 4th channel is impossible to access properly, but i suppose this will be adressed shortly with the new attribute field proposal is implemented?

Here's an example of the issue, tested on latest 3.0 alha build

Event Timeline

Laureline Amerio (lamerio) renamed this task from color alpha channel not yet implemented for attribute mix node to color type alpha channel not yet implemented (for attribute mix node).Sun, Jul 11, 6:14 PM
Laureline Amerio (lamerio) updated the task description. (Show Details)
Philipp Oeser (lichtwerk) changed the task status from Needs Triage to Needs Information from Developers.EditedMon, Jul 12, 3:06 PM

Well, I can confirm, it is always using ramp_blend in MA_RAMP_BLEND mode since the introduction in rBd72ec16e7072.

@Jacques Lucke (JacquesLucke) : should we be using blend_color_XXX instead? (like we do for CustomDataLayers)? These seem to support similar blend modes (and take alpha into account properly).
If so, I could look into making that change...

Wonder what shader and compositor nodes are using. Generally, I'm not opposed to using blend_color_xxx, haven't looked into it in detail now.

I think the change makes sense. It also means we could refactor the node to avoid the switch statement for the blend type on every single loop iteration. Are you still interested in doing that @Philipp Oeser (lichtwerk)? I could do it otherwise.

Actually, looking at this other set of functions, it seems they also don't do what we want, since they use the alpha value of the second color for mixing instead of a separate float factor.

@Hans Goudey (HooglyBoogly) : probably wont have time short-term, so go ahead.