MVP scope, honest version: full "form-correction notes from any video" needs a trained pose model, which is not bundled or fetched here. What's real and working: upload a short clip of a turn, this samples frames and tracks the dancer's real rotation via a brightness-centroid image-moment calculation, then reports actual turn count and rotation-speed consistency (the standard proxy for spotting/control quality) computed from your specific video -- not a canned response.
Best results on a clip with the dancer as the brightest object against a darker background (e.g. lit against a dim studio). This is a v1 signal, not a substitute for a teacher's eye.
Upload a clip and click Analyze.
Leave your email and we'll follow up directly -- no mailing list, no spam, just a real reply from a person.