Again, I really think there's only maybe three avatars where this matters, but it was bothering me and now it Works Gooder.
Instead of using the alpha mask directly, I now multiply it against a luminance version of the base image to create a new alpha mask, and then I use that. For cases where the original mask was properly only masking out white pixels, this will behave identically. For cases like Buck here, where Buck's mask was a little "anemic", you're going to mask out a lot less of the base image -- keeping in line with what CChat actually looks like.
For pixels that weren't fully white nor fully black, you'll get a little bit of semi-transparency, and that won't quite match the bitwise-anding of the original program, but it's gonna be just a few pixels on just a few poses on just a few avatars out of tens of thousands. It's fiiiine.

