blessing-video-weaverlisted

# Blessing Video Weaver — שזירת סרטוני ברכה לסרטון אחד Speech-first editing: the transcript drives the edit. Every clip's real content goes in (user's typical demand: "כל מה שהוסרט יהיה בסרטון"), but interleaved by THEME, not clip-after-clip. Proven end-to-end on a 9-clip / 10.7-min-raw → 10:19 final (2026-06-12). ## Pre-flight questions (AskUserQuestion, once) 1. **מוזיקה** — חגיגית רק בפתיח/מעברים | שקטה לכל האורך | שתיהן (ducking) | בלי 2. **אוויר מת** — מותר לחתוך שתיקות/גמגומים/סידור מצלמה? (תוכן אמיתי תמיד נשמר) Defaults you don't ask about: portrait 9:16 if sources are portrait; canvas = highest native resolution among sources (e.g. 576×1024@30); descriptive Hebrew output filename + a WhatsApp-friendly crf-25 copy. ## Pipeline (10 stages — templates/ has the proven scripts) The templates are the WORKED EXAMPLE from the 2026-06-12 production (9 siblings, Hebrew). Per project adapt: the `SEGS` table in cut_segments.py, the `CLIPS` map in normalize_all.py, the chapter/lower/behind variable lists in gfx/render_gfx.py, and the part lists in build_final.py. The ffmpeg flag choices, filter orders, and encoder settings are the hardened parts — keep them. | # | Stage | Tool / template | Output | |---|---|---|---| | 1 | Inventory | ffprobe every clip (W×H, dur, fps, **rotation side_data**) | plan table | | 2 | Transcribe | [templates/transcribe_all.py](templates/transcribe_all.py) — faster-whisper | `work/transcripts/*.json` | | 3 | Weaving script | manual authoring from