Video Podcast Maker

Recommended: Load Remotion Best Practices

This skill benefits from remotion-best-practices (not bundled) for the full Remotion pattern library. It is optional — minimum rules are below if absent.

• Pi: read the loaded skill at remotion-best-practices (listed in available skills).
• Claude Code: invoke remotion-best-practices skill/tool before proceeding.

Not installed? Get it from remotion-dev/skills (docs: remotion.dev/docs/ai/skills).

If remotion-best-practices is not installed, minimum rules: chromium must be available, always wrap 4K content in <Scale4K>, use <TransitionSeries> with linearTiming, and treat audio as the master clock.

Video Podcast Maker

Automated pipeline for 4K Bilibili horizontal knowledge videos from a topic. Coding agent + TTS backend + Remotion + FFmpeg.

Contents

• Bootstrap — prerequisites (run before Step 1)
• Execution Modes — Auto vs Interactive → references/workflow-script.md
• Regenerating an Existing Video — iterate on a finished video
• Workflow — the 11-step pipeline + phase-file pointers + mandatory stops
• Hard Rules — non-negotiable production constraints
• Audio-Master Clock & Sync
• Per-Video Layout
• Additional Resources — when to load each references/ file
• User Preferences
• Troubleshooting

Bootstrap

Resolve SKILL_DIR to the directory containing this SKILL.md:

• Pi: the agent knows the skill path from the loaded skill list — set SKILL_DIR to that directory before running commands.
• Claude Code: ${CLAUDE_SKILL_DIR} is auto-populated.

bashCopy code
SKILL_DIR="${SKILL_DIR:-${CLAUDE_SKILL_DIR}}"

# Prerequisites (CLIs + backend env vars)
python3 "${SKILL_DIR}/scripts/check_prereqs.py"

Updates flow through the skills CLI (npx skills update video-podcast-maker -g); direct git-clone installs use git pull per the README. This skill performs no update checks.

Prereqs failures — see README.md for setup. The check is backend-aware (resolves TTS_BACKEND env → user_prefs.json global.tts.backend → edge default), so only env vars required by the active backend are validated.

First video in a new project? Prefer reusing an existing Remotion project with node_modules/ already installed — creating a fresh project downloads ~2.2 GB of npm packages plus a 90 MB Chrome headless shell (one-time per project). If the user has a project from a previous video, use it. If a fresh project is necessary, run npm install in the background while you do Steps 1-4 (topic research and script writing).

All rendering goes into videos/{name}/ — every output.mp4, final_video.mp4, and thumbnail_*.png lands directly in the per-video directory. Never render to an out/ or dist/ directory; the --public-dir videos/{name}/ convention keeps everything self-contained.

TTS engine — two local backends, no external skill:

• edge (default) — free, no key, via edge-tts.
• azure — needs AZURE_SPEECH_KEY + AZURE_SPEECH_REGION (Microsoft Speech SDK).

Each synthesizes in-house (scripts/tts/backends/native.py) — pronunciation (display → spoken → back to display for subtitles) and phoneme application are built in. check_prereqs.py validates the active backend's env vars.

Multi-platform TTS? The former ttscn component skill that provided the 9-backend matrix is no longer a dependency of this skill. If you want those platforms, install Agents365-ai/ttsCN separately and call it directly — this skill ships only edge + azure.

Design Learning shortcut: If the user provides a reference video/image or asks to save/list/delete style profiles, see references/design-learning.md instead of running the workflow below.

Execution Modes

Detect Auto Mode (default) vs Interactive Mode at workflow start — the Auto-default decision table and per-request overrides are in references/workflow-script.md.

Regenerating an Existing Video

If videos/{name}/ already exists and the user is iterating on a finished or in-progress video, reuse that directory. Do NOT start a new project or a new videos/{newname}/.

Pick the smallest re-run for what actually changed:

Any re-run that changes what the viewer sees or hears re-enters the Step 8 gate: apply the change, let Studio hot-reload, and wait for a fresh explicit "render 4K" — the previous confirmation does not carry over. A script change shifts every downstream timestamp, so always regenerate timing.json through TTS — never hand-edit it. After any re-run, re-verify:

bashCopy code
python3 ${SKILL_DIR}/scripts/verify_output.py videos/{name}/

Workflow

Iterating on a finished video? If videos/{name}/ already exists, see Regenerating an Existing Video above for the minimal re-run — do NOT start at Step 1.

At Step 1 start, create one task per step in your agent's tracker. Mark in_progress on start, completed on finish. Files in videos/{name}/ are the durable record — if interrupted, inspect the directory to determine where to resume.

Mandatory stops (bold rows above):

• Step 8 — Studio review. MUST launch npx remotion studio and wait for user feedback before rendering. NEVER render 4K until the user explicitly confirms ("render 4K" / "render final"). A reply containing adjustment requests is not confirmation — apply the changes, let Studio hot-reload, and ask again. Every round of adjustments needs its own fresh confirmation before Step 9.
• Step 10 — verify_output.py. MUST pass before declaring the video done. Exit 0 = green; exit 2 = warnings still publishable. Auto-fixes common omissions (creates final_video.mp4 if missing). Validates publish info (title, description, tags, chapters) against the platform matrix — generate it in Steps 5.5 and 10.2. For machine-readable output add --format json.

Pre-render audit (recommended) — before Step 8:

bashCopy code
python3 ${SKILL_DIR}/scripts/audit_beat_sync.py <Video.tsx> <timing.json>

Flags beats that drift > 1.5s from narration.

Auto Mode: visual self-review. When running in Auto Mode (no user watching Studio), render 3-5 key frame stills before asking for render confirmation:

bashCopy code
npx remotion still src/remotion/index.ts <CompositionId> videos/{name}/_review_001.png --public-dir videos/{name}/ --frame=<midpoint_frame>

Pick frames at: hero title (~10% in), a dense section midpoint, and the outro. Read the stills back as images and run the design-guide.md and visual-taste.md checklists against actual rendered output. Catch overflow, contrast, and layout regressions before the 4K render. Delete _review_*.png after review.

Validation Checkpoints

Hard Rules

Visual minimums (text sizes, content width, safe zones, animation safety) live in references/design-guide.md. MUST load before Step 8.

Audio-Master Clock & Sync

Golden rules

1. Audio is the master clock. Every slide start, subtitle, chapter, and animation beat is derived from podcast_audio.wav and podcast_audio.srt.
2. Generate timing from TTS, not from text estimates. Pipeline: podcast.txt → generate_tts.py → podcast_audio.wav + podcast_audio.srt + timing.json → composition → render.
3. Never hand-write timing.json before audio exists. If you already have curated slides, run align_timing_from_srt.py to anchor them to the real SRT.
4. Compensate TransitionSeries overlap. TransitionSeries renders sum(section.duration_frames) - (N-1) * transitionFrames frames. Scale every section proportionally to keep the rendered length equal to timing.total_frames. Do not stuff all overlap frames into the first section. The corrected pattern is in templates/Video.tsx.

Mandatory sync checkpoints

If any checkpoint fails, stop. Do not publish.

Output Specs

Per-Video Layout

Copy code
project-root/                           # Remotion project root
├── src/remotion/                       # Remotion source (Root.tsx, compositions, index.ts)
├── videos/{video-name}/                # Per-video directory
│   ├── topic_definition.md             # Step 1
│   ├── topic_research.md               # Step 2
│   ├── podcast.txt                     # Step 4: narration script
│   ├── phonemes.json                   # Step 4.5: zh-CN pronunciation overrides
│   ├── assets/manifest.json            # Step 5: per-section asset registry
│   ├── publish_info.md                 # Step 10: title/description/tags
│   ├── podcast_audio.wav               # Step 7: TTS audio
│   ├── podcast_audio.srt               # Step 7: subtitles
│   ├── timing.json                     # Step 7: timeline (drives animations)
│   ├── thumbnail_*.png                 # Step 6
│   ├── output.mp4                      # Step 9: 4K render
│   ├── video_with_bgm.mp4              # Step 9: with BGM
│   ├── final_video.mp4                 # Step 10: final output
│   └── bgm.mp3                         # Background music
└── remotion.config.ts

--public-dir per video

Every Remotion command uses --public-dir videos/{name}/ — each video's assets stay in its own directory, enabling parallel renders:

bashCopy code
npx remotion studio src/remotion/index.ts --public-dir videos/{name}/
npx remotion render ... videos/{name}/output.mp4 --public-dir videos/{name}/ --video-bitrate 16M
npx remotion still ... videos/{name}/thumbnail.png --public-dir videos/{name}/

Naming

• Video name {video-name}: lowercase English, hyphen-separated (e.g. reference-manager-comparison)
• Section name {section}: lowercase English, underscore-separated, matches [SECTION:xxx]
• Thumbnails (16:9 AND 4:3 both required): thumbnail_remotion_16x9.png + thumbnail_remotion_4x3.png (or _ai_ prefix for AI-generated)

Additional Resources

Load on demand — do NOT load all at once:

All scripts are reachable through one dispatcher — start with python3 ${SKILL_DIR}/scripts/cli.py --help; full routes and envelope error codes: references/troubleshooting.md.

User Preferences

Mutable state (user_prefs.json, phonemes.json) lives in ~/.video-podcast-maker/ — safe from skill updates. Auto-migrated from the skill directory on first run. Run "show preferences" to view, or "set X Y" to change. Full commands: references/troubleshooting.md.

Troubleshooting

See references/troubleshooting.md on errors, BGM options, preference learning, design-learning issues.