MMAudio (8 s default) and HunyuanVideo-Foley generate audio for video, but production needs frame-accurate sync of footsteps, hits and ambience over minutes of picture. What objective sync metrics and human protocols exist, and can long-form results be stitched without audible seams or loudness jumps?
Why this is still open: Both projects report their own benchmarks; no independent long-form evaluation was found.