Atlas vs Shape splatter
Whenever I tried Atlas node it always makes sort of degraded alpha, kills silhouette details. The right is what I spend lots of time for by doing my own atlas node using either shape splatter or fx-map nodes. Both works just ok with silhouette details that usually important for small twigs, fallen leaves, grass etc.
The problem is both my own atlases , fx map and splatter based are slow as hell. Designer default Atlas is 3 times quicker. Considering you have to do it in twice of target resolution in all 3 ( including default Atlas) to have some anti-aliasing present.
So I wonder have anybody ever tried to fix default Atlas or it's a price for quickness we couldn't avoid?
Why it needs to chump off the alpha at all?
Replies
Assuming you're talking about the atlas splatter...
I had a bit of a poke around in the node and they're effectively converting the opacity to an alpha test - this appears to be done to facilitate using a flood-fill to generate information about the pattern.
my gut says that you could do a much more limited version of this node and still get good alpha with a pixel processor but honestly i think the better option at that point is to use an fx-map, even if it would be slower.
like i said in another thread - precomputing the position information is a good way to speed these things up. the less maths in the fx-map itself, the quicker it'll go
Thanks poopipe. I noticed default Atlas can't even do height blend properly . But looks like it does per-computing position scale and rotation for UV . So probably I could copy something useful from there. My guess it's the flood fill approach they use to detect pictures is main source of problem there. The flood fill never works right for something with complex silhouette and totally unnecessary. `
The solution to a lot of this sort of problem is for the developers to extend the blending capabilities of color fx-maps - ideally by exposing it to the user as a function.
I asked that many times for years . Last response was they unwilling to change the behaviour of basic nodes everything is built upon. So my guess it will never happen.
It's not like they can't alias the old node out like they did with the various blurs/autolevels/ambient occlusion nodes they've replaced over the years so that's a bad reason to not do it
In reality it's probably just a very niche issue that they aren't willing to put resources into implementing - while that would be disappointing it is a good reason.
The weird thing is they actually have MAx blending at least in drop down of color FX-map node but it never worked there.