A lens with several views of one place

Build one interactive web page: An empty log on a woodland walk, and a glass that shows who lives there, the same spot after dark, in winter and in spring. A field-guide label names whatever is under the glass. After dark the wood goes black and the glass is a torch; with sound on, it listens, and each bird grows louder as the glass nears it. Five registered pictures in one WebGL shader.

Set it on an invented business called Sampleton, with its own drawn or generated pictures and no one else's files. It suits: Any place with more to it than one moment shows: a nature reserve, a garden designer (now, in a year, in ten), a room furnished, empty or as a floor plan, a building by day and by night. Do not let it fail the way this kind of piece usually does: Too many views to load, labels nobody can read, or sound nobody asked for. Here only the walk and the first view load before the lens opens and the rest follow; every view is decoded before the glass may show it; and no audio is fetched or played until the visitor presses Listen.

It must work with the keyboard, on a phone, with reduced motion respected and no text under 12px. Tell me which libraries you chose and why.