Generate a SILENT B-roll clip of an existing avatar (and optional invented side-characters) DOING a demonstrative activity — not a talking head, no lip-sync — for an avatar reel. Given a short script, the avatar's reference image and the reel ratio (9:16 or 16:9), it authors ONE 6-panel storyboard sheet with the gpt-image-2 skill and animates that single sheet into an 8s 720p MUTED video with the seedance-2 skill. The avatar's voice is added later as voice-over by avatar-reel-composer (broll_source: existing), so these segments are intentionally silent and focus on what the avatar is DOING (POV, seen from behind, cooking, building, interacting with other characters), not …
