UGC talking-head video

Input: one character image (the frame you and the user picked from portrait-clone) and a script. Output: take_N.mp4 (raw) and take_N_captioned.mp4, plus the request JSON that made it.

Model: Seedance 2.5 through treg, reAPI route by default, PiAPI as the alternative (scripts/seedance_treg.py). No vendor key on this machine. Captions: ffmpeg + PIL (scripts/caption_burn.py). Both scripts are self-contained; read their docstrings before changing them.