All chapter text is now painted directly onto the canvas, frame by frame. It's not HTML over a video — it's text composited into the same pixel buffer as the video frame, sharing the same grain, vignette, shadow and depth. The text appears to live inside the room rather than float over it.
The only HTML overlay left is the "Add to ritual" button at the end (needs to be a real button you can click). Everything else — compass, spec strip, chapter headlines, body copy, even the loading state — is canvas-painted.
Frames re-extracted at full 1280×720 (was 720×405). Total payload now ~3.3MB.