Video Channels & Publishing

The Cinematic Audiobook pipeline renders upload-ready MP4s + thumbnails. Publishing them — to a YouTube channel per language (plus locale-appropriate alternatives), with curated metadata and consistent branding — is managed in a dedicated repo: video-channels.

It is the source of truth for what video exists, where it lives, what it says, and where it’s published. It stores metadata only — never the MP4 masters (those stay as reproducible build artifacts in data-cinematics).

The repo#

video-channels/
  platforms/accounts.yaml          # a channel per language + locale alternatives
  platforms/about.yaml             # paste-ready channel "About" text per language
  platforms/keywords.yaml          # channel keyword sets per language
  catalog/<book>/
    _book.yaml                     # shared book metadata (links, playlists, credits)
    c<n>.<lang>.yaml               # one record per (book, chapter, language) video
  templates/youtube-description.md # description shape + guidance
  schema/video-record.md           # field reference
  scripts/report.py                # "what we have / where / status" across the catalog
  brand/
    background.avif                # canonical Milky Way background (site-wide)
    brandkit.py                    # shared background compositor + SVG rasterizer
    generate_avatars.py            # 9 channel avatars (profile pictures)
    generate_banners.py            # 9 channel banners (channel art)
    avatars/<lang>.png             # 1024×1024
    banners/<lang>.png             # 2560×1440

Channel strategy#

One dedicated channel per site language, so each locale gets a coherent feed in its own language. YouTube is primary for most; where it’s blocked or not dominant, lead with locale platforms:

Channel registry#

English takes the bare brand; every other channel is @WheelOfHeaven{CODE} so the family is instantly recognizable. platforms/accounts.yaml is the source of truth — this table mirrors it. Flip status (planned → created → active) and fill channel_id as each channel comes online.

Each channel’s About text is in platforms/about.yaml (paste-ready per language, one line per paragraph; Elohim/Yahweh kept in the original spelling), and channel keywords in platforms/keywords.yaml (one localized set per language, under YouTube’s 500-character cap).

Site footer link (per i18n page)#

The YouTube link in the reading-site footer is per language: each localized page links to that language’s channel. It’s driven by a youtubeChannelUrl translation key in www.wheelofheaven.world/config.toml (one per language block), consumed by themes/bifrost/templates/partials/footer.html (and the social-icon partial + contact page) via trans(key="youtubeChannelUrl", lang=detected_lang).

A language whose channel isn’t created yet — and Hebrew (he), which has no channel (no Hebrew audio) — points at the EN flagship so no footer link 404s. When a channel goes live, set its youtubeChannelUrl to the real URL.

Channel branding#

Both generators share brand/brandkit.py, which composites over the canonical Milky Way background used site-wide (the same wheel-of-heaven-background image behind the OG cards / essentials, vendored as brand/background.avif) and rasterizes the bifrost logomark + wordmark SVGs (white via resvg) — so the kit stays in lockstep with the site. Each language gets a different horizontal pan across the galaxy, so the per-channel files stay distinct. Re-run either script if the logo, background, or palette changes.

Avatars (brand/generate_avatars.py → brand/avatars/<lang>.png, 1024×1024) — the logomark enlarged to fill the frame over the background. English is the unadorned flagship; every other channel carries a small lavender language code (DE/ES/FR/JA/KO/RU/ZH/TW). All content sits inside YouTube’s circular crop.

Banners (brand/generate_banners.py → brand/banners/<lang>.png, 2560×1440) — logomark + WHEEL OF HEAVEN wordmark + www.wheelofheaven.world, all inside the 1546×423 safe zone that shows on every device, with a soft edge vignette for the TV / desktop crops. Deliberately content-agnostic (no format tagline) so the channels can carry other content later.

The catalog#

One record per (book, chapter, language) — see schema/video-record.md for the full field list. It captures the render source (repo + path in data-cinematics, duration, resolution, fps — not the bytes), the curated publish metadata (title, description, tags, hashtags), and per-platform targets (account, status, URL).

Status vocabulary: planned → rendered → ready (metadata curated) → scheduled → published.

scripts/report.py prints a table across the catalog — status per target and whether the local MP4 exists — so gaps are obvious at a glance.

Description conventions#

Curated per language (translate the boilerplate; don’t machine-translate the lede). See templates/youtube-description.md. Two rules that matter for YouTube:

• One line per paragraph. YouTube preserves newlines literally, so never hard-wrap mid-paragraph in the YAML — one long line per paragraph, blank line between; YouTube soft-wraps it.
• Hashtags end the description (also mirrored in a hashtags: field). The first three surface above the title. Keep them ASCII (#Rael, not #Raël).

Title format: {book title} — Ch. {n}: {chapter_title} | Wheel of Heaven.

Publish workflow#

1. data-cinematics renders c{N}.{lang}.mp4 + c{N}.{lang}.thumb.jpg.
2. Add/refresh the catalog record here (python scripts/report.py shows gaps).
3. Curate the title/description/tags (per templates/).
4. Upload to the language’s channel with its avatar + banner; record the URL and flip the target’s status to published.

MP4s are uploaded manually today; a YouTube Data API uploader driven by the catalog is a natural future step.