Add things-worth-doing brand profile and short-form template fork

brands/things-worth-doing/brand.json

@@ -0,0 +1,36 @@
+{
+  "name": "Things Worth Doing",
+  "description": "Field guide / documentary / editorial-magazine aesthetic for a men's competence-and-doing content brand. Deliberately avoids neon, gamer, generic-AI-futuristic, and overly polished corporate looks — see the vault's Brand Strategy and Visual Identity sections.",
+  "version": "1.0.0",
+
+  "colors": {
+    "primary": "#B5651D",
+    "primaryLight": "#D98C4A",
+    "textDark": "#2A2620",
+    "textMedium": "#4A4438",
+    "textLight": "#8A8172",
+    "bgLight": "#F4EFE6",
+    "bgDark": "#171410",
+    "bgOverlay": "rgba(23, 20, 16, 0.85)",
+    "divider": "#3A352B",
+    "shadow": "rgba(0, 0, 0, 0.35)"
+  },
+
+  "fonts": {
+    "primary": "Arial, Helvetica, sans-serif",
+    "mono": "ui-monospace, Menlo, monospace"
+  },
+
+  "spacing": { "xs": 8, "sm": 16, "md": 24, "lg": 48, "xl": 80, "xxl": 120 },
+  "borderRadius": { "sm": 7, "md": 12, "lg": 16 },
+
+  "typography": {
+    "h1": { "size": 88, "weight": 700 },
+    "h2": { "size": 72, "weight": 700 },
+    "h3": { "size": 48, "weight": 700 },
+    "body": { "size": 44, "weight": 400 },
+    "label": { "size": 34, "weight": 500, "letterSpacing": 3 }
+  },
+
+  "_proposed_default": "Colors are a proposed default matching the 'field guide + documentary' visual identity (warm rust/amber accent, warm off-black background, warm off-white text) — not yet confirmed by Adrian. Change freely; nothing downstream depends on these specific hex values."
+}

brands/things-worth-doing/voice.json

@@ -0,0 +1,22 @@
+{
+  "voiceId": "cjVigY5qzO86Huf0OWal",
+  "description": "Voice Brief: calm, confident, conversational, knowledgeable, slightly understated (never announcer-register).",
+  "settings": {
+    "stability": 0.68,
+    "similarityBoost": 0.85,
+    "style": 0.15,
+    "useSpeakerBoost": true
+  },
+  "model": "eleven_multilingual_v2",
+  "qwen3": {
+    "speaker": "Ryan",
+    "language": "Auto",
+    "tone": "",
+    "instruct": "",
+    "clone": null
+  },
+  "sixtydb": {
+    "voiceId": "YOUR_VOICE_ID_HERE",
+    "settings": { "stability": 0.85, "similarity": 0.9, "speed": 1.0, "enhance": true }
+  }
+}

templates/things-worth-doing-short/CLAUDE.md

@@ -0,0 +1,74 @@
+# things-worth-doing-short — guidance for Claude
+
+Forked 2026-09-01 from `templates/concept-explainer-short` for the "Things Worth
+Doing" men's content brand ([[Men's UGC Project]] in the Obsidian vault, and
+[[Playbook - AI Agent Content Sprint (Things Worth Doing)]] for how a batch of
+scripts/shot lists gets produced upstream of this template). Same Python/moviepy
+pipeline, no Remotion, no npm — `gen_vo.py → gen_captions.py → build.py`, run
+**from the project directory**. Differences from the base template are below;
+everything not mentioned here works exactly like `concept-explainer-short`.
+
+## What's different from concept-explainer-short
+
+1. **One scene = one shot, not one narration beat.** The base template's own
+   guidance ("~15s pattern interrupt") assumes few, longer scenes. Things Worth
+   Doing's Production Standard requires visual movement every few seconds, so
+   TWD scenes run ~2-4s each — map each shot in the sprint playbook's Shot List
+   directly to one `scenes.json` scene, in order, one VO line per shot. A
+   28-second video is normal at 7-9 scenes here, not 4.
+2. **`visual` field gets a third real value: `"real"`.** `build.py` actually
+   decides treatment by file **extension**, not by the `visual` string — it's
+   documentation only. So a real filmed `.mp4` clip gets the exact same
+   boomerang-loop treatment an LTX clip would. Tag it `"real"` in scenes.json
+   purely so a human scanning the file can tell what still needs to be shot
+   vs. generated — `build.py` doesn't care.
+3. **Voice is ElevenLabs by default**, via `config.json → voice.brand:
+   "things-worth-doing"` (patched into this fork's `gen_vo.py` — the base
+   template didn't forward `--brand` to `tools/voiceover.py` at all, only
+   scene text and qwen3-specific fields). Real voiceId is still a placeholder
+   in `brands/things-worth-doing/voice.json` — see Gotchas.
+4. **Palette is the proposed "field guide + documentary" default** in
+   `config.json`/`brands/things-worth-doing/brand.json` — warm rust/olive
+   accents, warm off-black/off-white, deliberately not neon or AI-futuristic
+   (matches the brand's Visual Identity). Not yet confirmed by Adrian; safe to
+   change, nothing else depends on the specific hex values.
+5. **Real footage is expected, not optional.** TWD's playbook prioritizes real
+   product/filmed footage over AI generation wherever the shot needs to show
+   the actual product being used (see the playbook's visual hierarchy). Don't
+   default to generating everything with LTX/Ideogram just because this
+   template can — check each scene's per-shot production package first.
+
+## Working on a project copy (same as base template)
+
+1. Plan in `scenes.json` first — one scene per shot from the video's
+   production package (VO Script + Shot List sections), not from scratch.
+2. Hook discipline: scene 01 must earn the next few seconds — question or
+   tension immediately, no throat-clearing.
+3. `build.py` works at every stage — placeholders before assets, estimates
+   before VO, silent before audio. Render early, render often; show the user
+   intermediate renders rather than describing them.
+
+## Review checklist before calling a video done
+
+- Pull frames with ffmpeg at several timestamps and *look* at them: caption
+  collisions, asset crops, placeholder cards left in.
+- Check `gen_vo.py`'s per-scene wpm output; anything flagged FAST/SLOW that
+  `maxWpm` didn't catch needs a script edit or retake.
+- Run this video's package through the sprint playbook's QC gates and
+  Red-Team pass again post-render, not just at pre-production — a render can
+  surface problems (bad continuity, a shot that doesn't land) invisible in
+  text form.
+
+## Gotchas (this fork, so far)
+
+- **`brands/things-worth-doing/voice.json`'s `voiceId` is still
+  `"YOUR_VOICE_ID_HERE"`** — `tools/voiceover.py` will error rather than
+  silently using a default. Replace it with a real ElevenLabs voice ID before
+  running `gen_vo.py` for real.
+- Inherited from the base template: clone pacing follows the reference
+  recording, `voice.maxWpm` is a safety net not a fix; never burn whisper's
+  own transcription (captions force-align to script text on purpose); moviepy
+  2.x uses `with_*` methods and PIL for text.
+- Keep toolkit-level fixes (improving this *template*) in
+  `templates/things-worth-doing-short/`, not in a copied project under
+  `projects/` — same separation the base template's CLAUDE.md establishes.

templates/things-worth-doing-short/README.md

@@ -0,0 +1,126 @@
+# things-worth-doing-short
+
+Forked from `concept-explainer-short` for the "Things Worth Doing" brand — same
+mechanics documented below, but **see `CLAUDE.md` in this folder for what's
+actually different** (scene cadence, real-footage handling, ElevenLabs/brand
+wiring, palette). Read that first.
+
+---
+
+A 9:16 vertical **concept explainer** template — TikTok / Reels / YouTube Shorts
+style — built entirely from AI-generated parts and composed with a single Python
+build (no Remotion/Node). One concept, hook → explanation → payoff → CTA, with
+big burned karaoke captions for sound-off viewing.
+
+Everything is driven by **`scenes.json`**: each scene is narration text plus a
+visual asset. The pipeline turns that into per-scene cloned (or built-in) TTS,
+word-perfect caption timing, and an audio-anchored composite where timing drift
+is impossible.
+
+```
+scenes.json ──► gen_vo.py ──► gen_captions.py ──► build.py ──► out/short.mp4
+                  │                │
+                  ▼                ▼
+        audio/scenes/*.mp3   captions/words_*.json
+        vo_durations.json
+```
+
+## Quick start
+
+```bash
+cp -r templates/concept-explainer-short projects/my-short
+cd projects/my-short
+
+# 0. Renders immediately — placeholder cards, estimated timing, silent
+uv run build.py
+
+# 1. Write your scenes (see scenes.json for the example)
+# 2. Generate visuals (from the TOOLKIT ROOT — see "Visuals" below)
+# 3. Voiceover → captions → final render (from the project dir)
+uv run gen_vo.py
+uv run gen_captions.py        # needs: uv sync --extra whisper
+uv run build.py
+```
+
+## scenes.json
+
+```json
+{
+  "title": "Why Is the Sky Blue?",
+  "scenes": [
+    { "id": "01", "slug": "hook", "visual": "ltx",
+      "asset": "clips/01_hook.mp4",
+      "text": "Look up on a clear day..." }
+  ]
+}
+```
+
+- `text` — the narration, verbatim. Captions burn *this* text (whisper only
+  provides timing), so write it exactly as it should appear.
+- `asset` — relative path; the extension decides the treatment:
+  image → slow Ken Burns zoom, video → boomerang loop cut to the scene,
+  missing → gradient placeholder (so you can render at any stage).
+- `visual` — documentation of intent (`ideogram` / `ltx`); build.py goes by
+  the asset extension.
+
+**Pacing budget:** narration ÷ 2.4 ≈ seconds per scene. A 60s short is
+~140 words total; the 4-scene example is ~45s. Keep the hook under 3 seconds
+of setup. Platform limits: YouTube Shorts ≤ 3 min, Reels ≤ 3 min,
+TikTok ≤ 10 min — gen_vo.py and build.py print a warning past 3 minutes.
+
+## Visuals
+
+Generate from the **toolkit root**, save into the project's `images/` and
+`clips/`. The intended split (see CLAUDE.md "FLUX.2 vs Ideogram 4"):
+
+```bash
+# Text-bearing cards (Ideogram 4, JSON captions; 9:16 = --resolution 1440x2560)
+uv run tools/ideogram4.py --json caption.json --resolution 1440x2560 \
+    --output projects/my-short/images/02_scattering.png
+
+# Motion b-roll (LTX-2; 576x1024 scales exactly to 1080x1920)
+uv run tools/ltx2.py --width 576 --height 1024 --num-frames 161 \
+    --prompt "..." --output projects/my-short/clips/01_hook.mp4
+
+# Optional music bed (looped + ducked automatically if present)
+uv run tools/music_gen.py --prompt "..." --duration 120 \
+    --output projects/my-short/audio/music.mp3
+```
+
+Alternate cards and motion so the viewer gets a pattern interrupt every
+~15 seconds. Ideogram for anything with legible text; LTX for atmosphere.
+
+## Voice
+
+`config.json → voice` selects the narrator:
+
+```json
+"voice": {
+  "provider": "qwen3", "cloud": "modal", "maxWpm": 165,
+  "refAudio": "ref/my-voice.m4a",
+  "refText": "Exact transcript of the reference recording.",
+  "speaker": "Ryan", "tone": ""
+}
+```
+
+- **Clone**: set `refAudio` + `refText`. Use 12–25s of varied, full-sentence
+  speech *at narration pace* — the clone copies the reference's pace and
+  temperature cannot fix a rushed reference (see `/voice-clone`).
+- **Built-in**: leave `refAudio` empty; `speaker` + `tone` apply.
+- `maxWpm` is the pacing safety net: rushed takes are slowed in place with
+  pitch-preserving atempo. gen_vo.py prints per-scene wpm either way.
+- `provider: "elevenlabs"` also works (uses your configured voice).
+
+## Captions
+
+Karaoke pills, 1–3 words at a time, timed from whisper word timestamps but
+**force-aligned to your script text** so they're word-perfect. Tune position,
+size, and chunking in `config.json → captions`; defaults clear platform UI and
+the bottom rows of Ideogram list cards. Set `enabled: false` to skip.
+
+## Re-rendering
+
+Everything is idempotent and cached. Re-run `gen_vo.py --force` after script
+edits (then `gen_captions.py` again), or just `build.py` after swapping an
+asset or tweaking config. Timeline math always re-derives from
+`vo_durations.json` — there is nothing to manually re-time.