Only once, during installation, to download the models (~11 GB). After that, capture runs completely offline: your footage never leaves your computer. (One exception: if you later switch the Capture Backend to CPU Only or Non-NVIDIA, the first capture on that backend fetches a one-time extra component when online.)

Ordinary video, a phone clip, a webcam recording, or existing footage. For best results: one person, fully in frame, decent even lighting, fairly steady camera.

BlendCap focuses on a single subject per capture and tracks the main person in frame (it warns you if it sees several). For multiple characters, capture each person's clip separately. Support for true multi-person-capture is currently planned as a feature for a future release, but is still under development.

It depends on your GPU and the clip length. On a modern NVIDIA card it's roughly a third of a second per frame for the body; face capture is much lighter. The CPU Only backend should work on almost any machine but is much slower. Use Capture Skip to speed up captures in exchange for lower accuracy.

No, the paid build does not currently support macOS.

There's a CPU Only capture backend that runs almost anywhere, but it's much slower than GPU capture, and not recommended.

Rigify, Auto-Rig Pro, CloudRig, Mixamo skeletons and Mixamo control rigs via the bundled presets, and most other/custom rigs via the bone-map editor.

Yes. The BVH Library's Save BVH and Save FBX buttons export any capture for other software.

Yes. The BVH Library's Save BVH and Save FBX buttons export any capture for other software.

Finished takes go to your BVH Library; working data goes to a per-project cache. Both locations are configurable.