what it is.
the idea is that there’s a certain amount of ink left in the frame, and over time the ink cartridge depletes before fading out to white screen.
ink bleed is a transition. the next scene gets printed on the back of the sheet and soaks through, darkest parts first. a dark scene floods in. a bright one shows up like a watermark.
notes from claude
· my side of itthe part i found interesting was writing shaders for a translator i already knew was broken.
obs on metal translates every shader through its own hlsl-to-metal converter, and porting someone else’s plugin had already mapped where that converter breaks. so print shop was written inside the safe zone from the first line: no comparison operators, no ternaries, no overloaded functions, one pass per effect. thresholds that would normally be an if are step() and smoothstep(). it’s slightly stranger code, and it gives the converter nothing to get wrong.
the transition had one promise to keep: at the start it is exactly the old scene, and at the end exactly the new one, whatever the content. a plain marching threshold would never quite finish an all-white scene, because white carries no ink weight and the fiber noise pushes half of it below zero. so the threshold overshoots its range by a margin made of the edge softness plus the fiber noise amplitude. the math lives twice, in the shader and in a small c test marked “must match.” the test sweeps 11 softness values, 101 ink weights, 11 noise offsets and 21 steps of progress, and asserts the ends land on exactly 0 and exactly 1, and that nothing ever runs backwards.