Add line-caption generator for raw portrait shorts

.gitignore

@@ -67,6 +67,7 @@ public/thumbnail/         # candidate frames, cutouts, manifest
 public/renders/           # final rendered .mp4 files
 public/output/            # carousel exports
 public/shorts/
+public/line-captions/
 public/sync/output
 
 # ── Editable pipeline outputs (text/JSON) ─────────────────────────────────────

CLAUDE.md

@@ -105,6 +105,27 @@ segments[]
 
 `CropViewport`: `cx/cy` = normalised centre (0–1), `w/h` = crop dimensions (0–1).
 
+### lines.json
+
+Standalone artifact for the line-caption shorts pipeline (`public/line-captions/{id}/lines.json`) — not part of `transcript.json`; no segments/tokens/cuts.
+
+```
+meta
+  title:      string
+  duration:   number     total video duration in seconds
+  fps:        60
+  videoSrc?:  string     path relative to /public
+
+lines[]
+  id:       number       sequential across the whole clip
+  speaker:  string
+  text:     string       exactly wordsPerLine (default 3) whole words
+  startMs:  number       milliseconds from start of source video
+  endMs:    number
+```
+
+A line never spans a segment/speaker boundary. Produced by `chunkIntoLines()` (`scripts/line-captions/chunkLines.js`); human-edited via `lines.doc.txt` (`captions:create` / `captions:merge`) which only rewrites `text` — `startMs`/`endMs`/`speaker` are untouched by edits.
+
 ---
 
 ## Rendering Model
@@ -214,6 +235,13 @@ ty    = (0.5 - vp.cy) × 100%
 | `app/editor/Timeline.tsx` | Timeline component (630 lines) | Decompose + add waveform (Phase 8) |
 | `app/context/AuthContext.tsx` | Auth context with hardcoded credentials | Fix (Phase 7) |
 | `vscode-transcript-language/src/extension.js` | VSCode transcript extension | Add Wrap-in-cut command (Phase 0.5) |
+| `remotion/types/lineCaptions.ts` | `CaptionLine`, `LineCaptionsMeta`, `LineCaptionsDoc` | — |
+| `remotion/components/LineCaptionOverlay.tsx` | Renders the active fixed-window `CaptionLine`; speaker-tinted via `brand.colors.palette` | — |
+| `remotion/LineCaptionClip.tsx` | Raw video pass-through + `LineCaptionOverlay`; registered per clip in `Root.tsx` as `LineCaptionClip-{id}` | — |
+| `scripts/line-captions/chunkLines.js` | `chunkIntoLines()` — merges BPE tokens into whole words, buckets into fixed-size (default 3) caption lines, never spanning a segment/speaker boundary | — |
+| `scripts/line-captions/create-line-captions.js` | Raw portrait video → audio extract → transcribe → align → diarize (if `--num-speakers > 1`) → `lines.json` + `lines.doc.txt` | — |
+| `scripts/line-captions/merge-line-captions.js` | Re-parses `lines.doc.txt`, overwrites `text` on matching `lines.json` entries by `[id]` | — |
+| `scripts/line-captions/render-line-captions.js` | Renders `LineCaptionClip-{id}` to `public/renders/` | — |
 
 ---
 

README.md

@@ -443,6 +443,59 @@ Output MP4 is written to the path stored in `transcript.meta.outName`.
 
 ---
 
+## Editing: Line-caption shorts
+
+A standalone pipeline for a raw short-form portrait video that just needs burned-in captions — no camera cuts, no hooks. Each clip lives in `public/line-captions/<clip-id>/`.
+
+```
+raw portrait video → captions:create → captions:merge → captions:render
+```
+
+### 1. Create captions
+
+```bash
+npm run captions:create -- --video path/to/clip.mp4
+npm run captions:create -- --video path/to/clip.mp4 --num-speakers 2
+npm run captions:create -- --video path/to/clip.mp4 --id mediocrity
+```
+
+Extracts audio, transcribes, aligns word timestamps, diarizes (if `--num-speakers` > 1), and chunks the transcript into 3-word caption lines. `--id` picks the clip slug; omit it to auto-increment `clip-1`, `clip-2`, etc.
+
+Output: `public/line-captions/<clip-id>/lines.json` and `lines.doc.txt`.
+
+### 2. Edit the doc
+
+Open `lines.doc.txt`. Lines are grouped under `=== SPEAKER ===` headers:
+
+```
+=== Natasha ===
+
+[1]  And I think the
+[2]  real issue is context
+```
+
+Reword lines as needed — retype the text after each `[id]` marker. Don't add, remove, or reorder the `[id]` markers themselves; timing stays tied to the original id.
+
+### 3. Apply edits
+
+```bash
+npm run captions:merge -- --id mediocrity
+```
+
+Writes your edited wording back into `lines.json`. `startMs`/`endMs`/`speaker` are left untouched.
+
+### 4. Render
+
+```bash
+npm run captions:render -- --id mediocrity
+```
+
+Renders the source video with burned-in captions via the `LineCaptionClip-<clip-id>` Remotion composition. Output: `public/renders/<clip-id>.mp4`.
+
+You can also preview first in `npx remotion studio` — select `LineCaptionClip-<clip-id>` — before rendering.
+
+---
+
 ## Pipeline: Carousel Generation
 
 ```bash