Fifty new templates: end screens, timers, title cards and logo reveals
The library was strongest exactly where the competition is heaviest and thinnest where it already had pages that rank. Fifty templates later, that is the other way round.
What was actually missing
A commercial template pack ships about twenty-five lower thirds, twenty title cards, twenty logo reveals and ten end screens. This library had four lower thirds, three end screens and none at all of the other two — while the front page listed both titles and logo animations among the things you can make here.
Meanwhile the deepest shelves were subscribe animations, sponsor graphics and channel bars: twenty-five, twenty-nine and twenty-four. Those are the shelves every stock library on the internet also has. Being deepest where everyone is deep and thinnest where nobody is looking is a strange shape for a library to have, and it took a link-graph audit to notice it rather than a design review.
Ten lower thirds
A lower third is a text graphic pretending to be a design. Almost everything that makes one good is legibility, and the rest is where it sits — so these are one set of measurements carried by ten different constructions: a slab with a lit edge, a frosted pane, a hairline that draws itself, two brackets closing in on the words, offset blocks, a folded ribbon, a monospace caption, a gradient crossed once by a light sweep, an outline for footage too busy to sit a fill on, and an avatar plate.
Every plate is sized from the text it holds rather than to a round number, and the whole band sits a hundred and eighty pixels clear of the bottom edge — which is where YouTube draws its progress bar and controls the moment anybody touches the screen.
Ten end screens, built around what you do not own
YouTube draws its own elements over the last twenty seconds of a video: a card, and a subscribe button roughly a fifth of the frame wide. An end screen designed without them is an end screen with somebody else's button through the middle of it.
So every one of these keeps the right-hand third of the frame clear and puts its own content in the two thirds it can rely on. They run for ten seconds rather than six, because an end screen is watched while somebody decides whether to click something, and six is not long enough to decide in.
Ten timers and progress bars, which nobody sells editable
This is the shelf worth being pleased about. Stock libraries sell finished clips of countdowns and progress bars, and a finished clip counts down from whatever number the person who made it chose. A countdown is the graphic most obviously wrong when it is not yours, and it is the one nobody ships in a form you can change.
They sit at the top of the frame rather than along the bottom, because the bottom is where the player puts a progress bar of its own and a graphic there competes with the thing it looks exactly like. And every fill moves linearly: a bar that eases is a bar lying about how much time is left.
Ten title cards and ten logo reveals
A title card is the only kind of graphic here that takes the whole frame. For a second or two it is the video, which is why every one of these leaves — the hold is about a second and a half, long enough to read twice and short enough that nobody reaches for the scrubber.
The logo reveals are shorter still. Four seconds at most, because an intro runs before anybody has decided to stay and every extra second is measurably people leaving. The mark in each one is an image layer called Logo: drop your own file on it, or import your channel and it fills with your own picture. That is the part a baked clip cannot do.
How they were checked
Writing a template is writing coordinates, and coordinates are exactly the thing that reads as correct in source and wrong on screen. The first pass of the lower thirds type-checked, passed a geometry test written for it, and still had an accent bar painted over by the plate that was supposed to sit behind it, a scrim that ended mid-word because the width estimator is calibrated for Inter and that one design uses a monospace face, and a ribbon dissolving under its own second line.
None of those are visible in code. So each pack was rendered through the same canvas the editor uses and looked at, and the ones that were wrong were fixed and looked at again. The rig that does it is in the repository now, because the next fifty will need it too.
One fault was worth a permanent test rather than a fix. Two keyframes on the same moment of the same property ask the renderer to interpolate across a segment of no length — it does not raise anything, it simply never reports the frame as drawn, and the template hangs forever. It was written by hand twice in one afternoon and was invisible to every check that existed. Every template in the library is now checked for it.