Building a 350-item motion catalog: licensing, attribution, and what we learned
Most of the catalog is adapted from an Apache-licensed registry and one MIT-licensed component set. Here is what that actually obligates us to do, and what it doesn't.
The catalog is 350 motion components and blocks, and almost none of it started as original work in the sense that matters here — the majority is adapted from an Apache 2.0 licensed registry, with a smaller set of components drawn from an MIT-licensed project. Both grants are permissive, and both come with real obligations that are worth stating plainly rather than treating as a footnote.
What Apache 2.0 actually requires
Section 4 of the Apache license binds derivative works — and heavy modification is what makes something a derivative work, not an exemption from the license covering it. The obligation is specific: retain the original copyright notice, and state that you modified the files. That's it. It does not require publishing your changes, and it does not transfer any claim over your own additions.
We did modify heavily — the source items were standalone HTML documents driving a paused GSAP timeline for a host page to seek; every one of the 350 ported components now runs on Remotion's frame clock instead, with DOM lookups scoped to the component root and CSS namespaced so it doesn't resize the page around it. That's real, substantive rewriting. It still doesn't change what Section 4 asks for: every adapted source file carries a header stating what was changed, and the attribution page carries the full changelog and the copyright notice.
Where the actual line is
Copyright protects expression, not technique. "Letters descend from above in a staircase with zero blur" describes a common motion-design move that predates any registry involved — nobody owns that idea, and two independent implementations would describe it almost identically because there aren't many ways to say it. The place a claim can plausibly sit is narrower: the literal source code (which we rewrote), and the specific taxonomy — categories, tags, the selection metadata that makes the catalog searchable by job and family rather than by keyword. That metadata is derived, not independently authored, which is exactly why the attribution notice exists rather than a claim that everything here is original.
Binary assets are a separate question
A smaller set of items ship binary assets — fonts, 3D models, texture files — copied alongside the source code they support. Those assets carry their own upstream terms, independent of the Apache grant on the code, and the honest state of that work today is: identified, catalogued, not yet fully cleared. Fonts are the highest-risk category, since a desktop font license frequently excludes web or video embedding even when the code around it is fully open. This is active, ongoing work rather than a solved problem, and the attribution page will get more specific as it's cleared item by item.
The pragmatic takeaway
An Apache or MIT grant is not a reason to skip attribution — it's the reason attribution has a specific, satisfiable shape instead of an open-ended one. Do the two things the license actually asks for, keep the notice traveling with the work, and the rest of the obligation is met. The part that takes ongoing care is the part no license grant covers at all: the assets riding alongside the code.