B-Roll Pack

Generate a b-roll pack that cuts together like one shoot. The discipline: ONE master look block, repeated verbatim in every clip; only the subject line changes.

What the b-roll supports: [THE A-ROLL/PROJECT: "a talking-head video about remote work" / "my SaaS launch video" / "a coffee brand's social content"]
The cutaway moments I need to cover: [LIST THE SCRIPT BEATS THAT NEED VISUALS, e.g. "mornings/routine, focus/deep work, the product on screen, city energy, winding down"]

MASTER LOOK BLOCK (write once, paste into every clip prompt unchanged):
[LENS + GRADE + LIGHT + ENERGY, e.g. "Shot on 35mm, shallow depth of field, muted warm grade with soft film grain, natural window light, calm unhurried motion": name a lens and a grade; 'cinematic' alone steers nothing.]

THE PACK: one clip per beat, each prompt = [SUBJECT + ONE MOTION] + [MASTER LOOK BLOCK] + [AUDIO LINE]:
1. [BEAT 1]: [e.g. "Close-up, hands wrapping around a steaming mug on a wooden counter, steam curling"]; camera: [static / slow push-in].
2. [BEAT 2]: [e.g. "Over-shoulder of hands typing on a laptop, screen softly out of focus"]; camera: [slow lateral drift].
3. [BEAT 3]: [e.g. "The city street below through a rain-flecked window, umbrellas passing"]; camera: [static].
[CONTINUE PER BEAT, standard coverage categories if I need filler: hands working / establishing wide / product macro / nature-atmosphere / transit motion]

Motion rules for edit-friendliness:
- One subject, one motion per clip; camera static or one slow constant-speed move (push-in, drift, pan): constant velocity trims, loops, and speed-ramps cleanly; accelerating or turning moves fight the edit.
- 4-8 seconds each, composed with breathing room for crops to vertical.
- Nothing that reads as a scene with a beginning and end: b-roll is texture, not story.

Audio line per clip (this is what makes generated b-roll feel real): SFX + ambience matched to the shot ["the mug's soft ceramic set-down, distant kitchen hum" / "muffled rain, faint street noise"]. No music: the edit owns the music.

No people's faces unless listed (faces date the pack and limit reuse); hands and backs-of-heads travel better.