AGENT SKILL

shorts-factory

Use when producing short-form vertical or square videos (Shorts, Reels, TikTok) end to end from a topic -- script, narration, alignment, stock b-roll, scene plan, render, and ffmpeg composite with captions. Assembles tts + extensions/asr + extensions/stock-footage + animation-direction/remotion-render into a MoneyPrinterTurbo-equivalent pipeline.

License 许可
MIT
最近更新
2026-11-15
Stars
★ 0
Forks
1

01 前言 Front Matter

name
shorts-factory
triggers
[ "user", "model" ]
description
Use when producing short-form vertical or square videos (Shorts, Reels, TikTok) end to end from a topic -- script, narration, alignment, stock b-roll, scene plan, render, and ffmpeg composite with captions. Assembles tts + extensions/asr + extensions/stock-footage + animation-direction/remotion-render into a MoneyPrinterTurbo-equivalent pipeline.

03 SKILL.md 正文

Shorts Factory

Topic in, captioned short video out. No video editor, no paid TTS/ASR. Reuses bundle pieces; the only new moving part is extensions/stock-footage/fetch.py for b-roll clips.

Pipeline

topic -> script -> narration WAVs -> transcript segments
      -> scene-plan v2 (+ stock clips) -> render -> ffmpeg composite

1. Script (agent prompt pattern)

Prompt the agent for a spoken script, one line per caption beat:

  • Target 45-90 words for a 30-60s short (160 WPM read time).
  • One sentence per line; each line becomes a narration segment and a caption. No stage directions, no headers.
  • End with a hook or call to action line.

2. Narration (tts skill)

One WAV per script line via extensions/voicebox/wrapper.py:

python extensions/voicebox/wrapper.py generate \
    --text "<line>" --voice <id> --out narration/line_01.wav

Default voicebox backend, or --backend voicestudio / VB_BACKEND=voicestudio for the VoiceStudio app on 127.0.0.1:3900. doctor first; status: missing means tell the user to start the backend. Concatenate WAVs (ffmpeg concat demuxer) into narration.wav recording each line's cumulative start offset.

3. Alignment (extensions/asr)

python extensions/asr/transcribe.py narration.wav --model base
# -> {"ok": true, "segments": [{start, end, text}], ...}

Segments anchor captions and scene sync_ranges. Requires extensions/asr/bootstrap.py to have fetched whisper.cpp + model (explicit fetch, never silent). Missing binaries -> ok:false + hint.

4. B-roll (extensions/stock-footage)

python extensions/stock-footage/fetch.py search --query "<kw>" \
    --provider pexels --orientation portrait --per-page 10
python extensions/stock-footage/fetch.py download --provider pexels \
    --id <id> --out clips/clip_01.mp4

Keys from PEXELS_API_KEY / PIXABAY_API_KEY; absent -> {"ok": false, "status": "missing", "hint": ...} with zero network calls. Download picks the smallest file meeting --min-width (default 720). Stock clips are optional: skip this stage and let the render stage produce all visuals when no keys are set.

5. Scene plan (scene-plan v2)

Reuse the scene-plan v2 schema from skills/animation-direction (craft-bridge scene_manifest v1 + motion_spec + sync_range):

{"scenes": [{"id": "s1", "sync_range": [0.0, 3.2],
  "visual": {"type": "kinetic_text", "props": {...}},
  "motion_spec": {"easing": "easeOut", "enter": "fade",
                  "exit": "fade", "enter_s": 0.3, "exit_s": 0.3},
  "broll": "clips/clip_01.mp4"}]}

One scene per transcript segment/beat. broll is a local downloaded clip path; visual covers text/diagram beats the same way as animation-direction. Validate with extensions/remotion-render/plan_check.py and get plan approval before rendering (hard gate, same as animation-direction).

6. Render

python extensions/remotion-render/render.py plan.json --workdir out/

Remotion renders the scene visuals/overlays. Full 3D scenes route to operate-blender; pure text+b-roll plans may skip render entirely and go straight to composite.

7. Composite (ffmpeg)

# 9:16 shorts (1080x1920); use 1920x1080 for 16:9, 1080x1080 for 1:1
ffmpeg -i render.mp4 -i narration.wav \
  -vf "scale=1080:1920:force_original_aspect_ratio=decrease,\
pad=1080:1920:(ow-iw)/2:(oh-ih)/2,subtitles=captions.srt:force_style='Fontsize=14'" \
  -c:v libx264 -c:a aac -shortest short.mp4

Build captions.srt from ASR segments. Pick aspect per platform: 9:16 Shorts/Reels/TikTok (default), 1:1 feed, 16:9 YouTube.

Example flow

W=extensions/voicebox/wrapper.py
python $W doctor                                    # backend up?
python $W generate --text "Three facts about octopi." \
    --voice af_heart --out n/line_01.wav            # per line...
ffmpeg -f concat -safe 0 -i wavlist.txt -c copy narration.wav
python extensions/asr/transcribe.py narration.wav > transcript.json
python extensions/stock-footage/fetch.py search \
    --query "octopus ocean" --orientation portrait > hits.json
python extensions/stock-footage/fetch.py download --id 857195 \
    --out clips/c1.mp4
# agent writes plan.json (scene-plan v2) from transcript + hits
python extensions/remotion-render/plan_check.py plan.json
python extensions/remotion-render/render.py plan.json --workdir out/
ffmpeg -i out/out.mp4 -i narration.wav -vf "scale=1080:1920,subtitles=c.srt" \
    -c:v libx264 -c:a aac -shortest short.mp4

Typed errors per stage

Stage Failure Envelope
tts backend down status:"missing" + start hint
tts bad voice/engine status:"error", error detail
asr missing binary/model ok:false + bootstrap hint
asr fetch/decoder failure ok:false, error:"fetch_error:*"
stock no API key status:"missing" + hint, no network
stock HTTP/timeout status:"error", error:"http_error:<code>"
render plan invalid plan_check.py findings, gate blocks
composite missing input ffmpeg nonzero exit; surface stderr

Never fabricate: no fake audio, no fake transcript, no fake clips. Every stage degrades to a typed error, not a traceback.

Non-goals

  • No auto-publishing or platform upload (YouTube/TikTok APIs).
  • No batch mode, watch folder, or scheduling daemon.
  • No paid TTS/ASR/stock services; no API keys beyond Pexels/Pixabay.
  • No long-form video; 3 minutes max is a hard ceiling.

Boundaries

  • Localhost services only (voicebox, VoiceStudio). Keys stay in env, never logged or written to plans.
  • Stock clips carry provider license terms; keep hits[].url for attribution.
  • Plan approval gate is mandatory before render.py on non-trivial plans, same as animation-direction.

04 同仓库更多 Skills

Leostruka
Leostruka/devin-bundle

montador-de-ideias

Use when assembling, expanding, or elevating a raw project idea or initial prompt into an ultra-detailed, first-principles dossier and compiled master prompt with journalistic OSINT market research, Tech Lead and Staff Architect justifications, atomic and modular hardening/optimization, and high-impact contact surfaces.

★ 0
Leostruka
Leostruka/devin-bundle

git-github-adaptive-engineering

Use when any task touches git history, GitHub issues/PRs/releases/CI, repo governance, or branch and merge decisions. Runs the Fase Zero situational-awareness battery, picks one of five delivery flows via the adaptive decision engine, and enforces commit/PR information-density standards. Orchestrates the gh, git-workflows, github-actions, pr, and debug/finish skills.

★ 0
Leostruka
Leostruka/devin-bundle

adaptive-agenda-calendar-routine-engine

Use when planning or executing work across hours, days, weeks, or sprints. Turns the calendar into a strict time-blocked agenda (HH:MM-HH:MM blocks carrying a routine profile, expected inputs, and a handoff output artifact), runs the 5-stage cognitive flow (Radar and Hunt -> Triage plus Steve Jobs conception -> workbench manifest -> >=32-cycle loop -> Close plus Open-Minded Scrum), and unfolds macro calendar milestones into micro agenda blocks through extensions/gcal/wrapper.py.

★ 0
Leostruka
Leostruka/devin-bundle

demo

Use when demonstrating a skill in a fixture project.

★ 0
Leostruka
Leostruka/devin-bundle

duplicate-skill

Use when testing duplicate skill name detection.

★ 0
Leostruka
Leostruka/devin-bundle

duplicate-skill

Use when testing duplicate skill name detection.

★ 0