Great breakdown of the segment-based approach. The prompt structure you described (scene description, key actions, starting/ending pose, timestamps) lines up with what works best — subject first, then camera, then lighting/audio cues.
That's the actual product page where you can try the model. It launched globally on Feb 12.
For a structured walkthrough (account setup, prompt structure, the @ reference system), I put together a guide here: https://seedanceguide.com/tutorial
I documented 25 tested prompts following this pattern if anyone wants ready-to-use templates: https://seedanceguide.com/blog/best-seedance-2-0-prompts