Zman2087/faceless-studio

Asset library

One register per channel in channels/<slug>/assets/library.json (files in assets/files/), plus channels/_shared/ for things every channel can use. Channel folders are git-ignored by default, so the library stays private. Everything goes through scripts/assets.mjs:

node scripts/assets.mjs seed  <slug>                                   # after onboarding: imports face anchor, sheet, location stills
node scripts/assets.mjs add   <slug> <file|url> --kind broll --tags kitchen,morning --licence pexels --source <page URL>
node scripts/assets.mjs find  <slug> --kind broll --tags kitchen --min-dur 3 --unused-days 30
node scripts/assets.mjs use   <slug> a-007 --post p-012
node scripts/assets.mjs check <slug>

Kinds and tags

clip (AI character clips we generated), broll, still (character/location images), photo, music, sfx, voice (approved voice takes), font, logo-free-graphic. Tags: lowercase, concrete, searchable — the place (kitchen), the mood (calm), the subject (coffee), the pillar key from settings. 2–5 per asset.

Licences — the only rule that matters

--licence	use it?	credit
own	yes — we generated it (Higgsfield, ElevenLabs) or the owner made it	no
cc0, pexels, pixabay	yes	no (still record --source)
unsplash	stills only	no
cc-by	yes	required: --credit "Photo: Name, CC BY 4.0" → put it in the shot's credit
licensed	yes — paid; put the order/licence ref in --note	as the licence says
unknown	never published — find hides it, use refuses it	—

Never add: anything with a brand logo or a real person's face, screenshots of real chats or posts, music ripped from TikTok/YouTube/Spotify, "free" sites that don't state a licence. When the owner gives a file without a source, add it as unknown and ask where it came from.

Where assets come from

1. Ours first: every Higgsfield clip and still the producer makes is own; good unused takes are free b-roll later (--kind clip --job-id …).
2. Free stock with a clear licence: Pexels / Pixabay video & photos, CC0 libraries. Save the page URL as --source.
3. Music: the app's own sound library when posting is safest (trending sounds are added in-app, not baked in). Baked-in beds only with own/cc0/licensed; keep them at music_volume 0.06–0.12.
4. Generate what's missing (Higgsfield generate_image/generate_video, announce credits first) and add it as own.

How the pipeline uses it

• Writer / video-script: find --unused-days 30 for b-roll so the feed doesn't repeat itself; least-used comes first.
• Producer: after a post is produced, add its clips (--kind clip --licence own --job-id) and use every asset that appeared in it.
• Before delivering any video that used non-own assets: check must print CHECK OK, and every cc-by asset's credit is on its shot.
• In the cloud sandbox the library isn't checked out: pass assets as their url (or upload the file with media_upload and use that URL).