Face swap or avatar talking video: which Sume endpoint?

Pick Beta face swap when a source video already exists and you want a ready avatar's face applied to it; pick the avatar talking video when you have a script and want the avatar to perform it. Both need an avatar first, and the face swap docs say it is not for script-driven talking video.

What does each endpoint take?

Both use a ready avatar referenced by avatar_handle. Face swap adds a public HTTPS video_url and a required quality. Avatar video adds exactly one of script or video_inputs, with optional product image, scene, quality, aspect_ratio and captions.

When is face swap the right one?

When you have footage: a short clip with usable audio, with the avatar's face on it. The docs also say it is not the old consumer-product /face-swap route; use only the Developer API path above.

When is the talking video the right one?

When you start from words. Avatar video reads a script, supports multi-scene plans with silence beats, and can burn captions in. Requests outside 4-60 estimated seconds are rejected, so long scripts must be split. Details in the avatar video docs.

Do I need to make the avatar in both cases?

Yes. Create it once from a prompt, traits or photo in the avatar docs and reuse the handle. Getting consent for any real person's face is on you, since the docs show no consent step.