xianyu110/ecommerce-video-skills / skills/ffmpeg-auto-assemble
ffmpeg-auto-assemble
Verifiedffmpeg 自动拼接成片:一条命令把 storyboard.json(商品图/视频片段 + 口播稿)渲染成带配音、字幕花字、转场、BGM 自动闪避、响度标准化的 9:16 MP4,纯本地运行。Use when the user wants to assemble/edit/render a product video, 自动剪辑, 一键成片, 拼接视频, 图片转视频, Ken Burns, add BGM/subtitles/voice-over with ffmpeg, or batch-produce short videos without an editor.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.0 KB |
| scripts/assemble.py | script | 14.5 KB |
| scripts/make_bgm.py | script | 2.8 KB |
| scripts/make_cards.py | script | 8.0 KB |
| scripts/make_subs.py | script | 5.6 KB |
| scripts/tts_edge.py | script | 4.5 KB |
scripts/tts_edge.py
View on GitHub#!/usr/bin/env python3
"""AI voice-over with edge-tts (free Microsoft Edge online voices).
Usage:
python tts_edge.py "夏天带冰水出门,不到中午就变温?" -o out/l00.wav
python tts_edge.py --storyboard storyboard.json --outdir build/tts # one wav per shot line
python tts_edge.py --list zh-CN # list voices for a locale
Each line is synthesised to mp3, leading/trailing silence trimmed, converted to
48 kHz mono WAV, and its duration written to <outdir>/durations.json.
Results are cached by (text, voice, rate, pitch) so re-runs are instant.
Needs: pip install edge-tts ; ffmpeg on PATH ; network access to the Edge TTS service.
"""
import argparse, asyncio, hashlib, json, os, subprocess, sys
DEFAULT_VOICE = "zh-CN-YunxiNeural"
TRIM = ("silenceremove=start_periods=1:start_threshold=-45dB,areverse,"
"silenceremove=start_periods=1:start_threshold=-45dB,areverse")
def probe_duration(path):
out = subprocess.check_output(["ffprobe", "-v", "error", "-show_entries", "format=duration",
"-of", "csv=p=0", path])
return float(out.strip())
async def _synth(text, voice, rate, pitch, mp3):
import edge_tts # imported lazily so --help works without it
await edge_tts.Communicate(text, voice, rate=rate, pitch=pitch).save(mp3)
def synth_line(text, out_wav, voice=DEFAULT_VOICE, rate="+0%", pitch="+0Hz", retries=3):
"""Synthesise one line to a trimmed 48 kHz mono wav. Returns duration in seconds."""
os.makedirs(os.path.dirname(os.path.abspath(out_wav)), exist_ok=True)
key = hashlib.sha1(f"{text}|{voice}|{rate}|{pitch}".encode()).hexdigest()[:12]
stamp = out_wav + ".key"
if os.path.exists(out_wav) and os.path.exists(stamp) and open(stamp).read() == key:
return probe_duration(out_wav)
mp3 = out_wav[:-4] + ".mp3"
last = None
for attempt in range(retries):
try:
asyncio.run(_synth(text, voice, rate, pitch, mp3))
if os.path.getsize(mp3) > 0:
break
except Exception as e: # network hiccups are common; retry
last = e
else:
raise RuntimeError(f"edge-tts failed for {text!r}: {last}")
subprocess.run(["ffmpeg", "-y", "-v", "error", "-i", mp3, "-af", TRIM, "-ar", "48000", "-ac", "1", out_wav],
check=True)
open(stamp, "w").write(key)
return probe_duration(out_wav)
def synth_storyboard(sb, outdir, voice=None, rate=None, field="line"):
voice = voice or sb.get("voice", DEFAULT_VOICE)
rate = rate or sb.get("rate", "+0%")
durs = []
for i, shot in enumerate(sb["shots"]):
text = shot[field] if field in shot else shot.get("line")
if not text:
durs.append(0.0)
continue
d = synth_line(text, os.path.join(outdir, f"l{i:02d}.wav"), shot.get("voice", voice),
shot.get("rate", rate), shot.get("pitch", sb.get("pitch", "+0Hz")))
durs.append(round(d, 3))
print(f" tts {i:02d} {d:5.2f}s {text}")
json.dump(durs, open(os.path.join(outdir, "durations.json"), "w"))
return durs
async def _list(prefix):
import edge_tts
for v in await edge_tts.list_voices():
if v["ShortName"].startswith(prefix):
print(f'{v["ShortName"]:<34} {v["Gender"]:<7} {", ".join(v.get("VoiceTag", {}).get("VoicePersonalities", []))}')
def main():
ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
ap.add_argument("text", nargs="?")
ap.add_argument("-o", "--out", default="out/voice.wav")
ap.add_argument("--storyboard")
ap.add_argument("--outdir", default="build/tts")
ap.add_argument("--field", default="line", help="storyboard field to read (e.g. line_en)")
ap.add_argument("--voice")
ap.add_argument("--rate", help='e.g. "+10%%"')
ap.add_argument("--pitch", default="+0Hz")
ap.add_argument("--list", metavar="LOCALE", help="list voices whose name starts with LOCALE, e.g. zh-CN, en-US, ja-JP")
a = ap.parse_args()
if a.list is not None:
asyncio.run(_list(a.list)); return
if a.storyboard:
sb = json.load(open(a.storyboard, encoding="utf-8"))
durs = synth_storyboard(sb, a.outdir, a.voice, a.rate, a.field)
print(f"total voice {sum(durs):.2f}s -> {a.outdir}/durations.json")
elif a.text:
d = synth_line(a.text, a.out, a.voice or DEFAULT_VOICE, a.rate or "+0%", a.pitch)
print(f"{a.out} {d:.2f}s")
else:
ap.print_help(); sys.exit(1)
if __name__ == "__main__":
main()
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |