yoyo-video-clipping-workflow
Skill9 starsChanged 7 months agoMITView on GitHub
悠悠有鸽竞品视频精剪流水线 - Whisper转录 + AI识别带货节点 + 精剪切片
- Installs packages
In this repository19 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- douyin-video-acquisitionSkill
- ecommerce-short-video-hook-researchSkill
- ecommerce-video-clip-to-shortformSkill
- ecommerce-video-clip-workflowSkill
- ecommerce-video-discoverySkill
- ecommerce-video-highlightsSkill
- ecommerce-video-publishingSkill
- ecommerce-visual-clip-scanningSkill
- error-attribution-analysisSkill
- hermes-routerSkill
- jianying-editorSkill
- jianying-editor-skillSkill
- neirong-fuoliSkill
- speech-quality-reviewSkill
- system-health-monitorSkill
- task-delivery-scoringSkill
- whisper-video-clipping-workflowSkill
The file
Files
悠悠有鸽 - 竞品视频精剪流水线
概述
从悠悠有鸽(香港YoYo)直播竞品视频中,通过 Whisper 转录 → AI分析识别带货节点 → 剪辑精剪切片的完整流水线。
环境准备
Python 环境
- AutoClip 需要 Python 3.11+(系统默认 3.9 不支持
type |语法) - 使用
~/.hermes-guardian/venv/bin/python3.11 - 依赖安装:
pip install fastapi uvicorn python-multipart dashscope==1.23.5 \ openai pydub pysrt pydantic python-dotenv aiofiles \ requests aiohttp cryptography yt-dlp watchfiles whisper
Whisper 模型
- 模型缓存位置:
~/.cache/whisper/ - 已下载:
base.pt,large-v3-turbo.pt - whisper CLI 位置:
~/.hermes-guardian/venv/bin/whisper
流水线步骤
Step 1:提取音频(ffmpeg)
视频文件大(4GB+),需要先提取音频再转录。
# 提取音频(WAV格式,单声道16kHz)
ffmpeg -y -i "INPUT.mp4" \
-ac 1 -ar 16000 -c:a pcm_s16le \
-t 600 \ # 可选:限制时长(秒)
"output.wav"
# 快速测试:只提取前1分钟
ffmpeg -y -i "INPUT.mp4" \
-ac 1 -ar 16000 -c:a pcm_s16le \
-t 60 \
"output_1min.wav"Step 2:Whisper 转录
# 创建输出目录
mkdir -p /tmp/whisper_output
# 使用 tiny 模型(最快,用于测试)
whisper "audio.wav" \
--model tiny \
--language zh \
--output_format srt \
--output_dir /tmp/whisper_output
# 使用 base 模型(平衡速度与精度)
whisper "audio.wav" \
--model base \
--language zh \
--output_format srt \
--output_dir /tmp/whisper_output性能参考(Mac M2 CPU):
- 1分钟音频:约2分钟(tiny模型)
- 10分钟音频:转录超时(约20分钟+)
base模型比tiny慢约3-5倍- 建议分批处理或使用 GPU
Step 3:AutoClip 后端
# 启动后端(需要 Python 3.11)
cd ~/autoclip_mvp
mkdir -p uploads outputs logs # 首次需要创建
PYTHONPATH=~/autoclip_mvp \
~/.hermes-guardian/venv/bin/python3.11 \
backend_server.py
# 健康检查
curl http://localhost:8000/healthStep 4:AutoClip 前端
cd ~/autoclip_mvp/frontend
npm install # 首次需要
npm run dev # 启动后访问 http://localhost:3000测试视频路径
/Users/yehya/Downloads/竞品视频/香港🇭🇰YoYo20260425201954.mp4
- 时长:100分29秒
- 大小:4.3GB
- 适合测试完整流程悠悠有鸽转录样本
[00:00.000 → 00:03.720] 这个门店去给来去做这么一个直播
[00:03.720 → 00:04.360] 我们最近的话呢
[00:04.360 → 00:08.600] 也是有很多的一些上心
[00:08.600 → 00:09.960] 大部分的话都是断秀
[00:09.960 → 00:12.600] 那么最近也是一直后台很多老粉
...→ 主播正在介绍新品上架,引导观众关注
AutoClip 输出说明(重要)
AutoClip 识别的是话题级片段(3-10分钟),不是最终精剪!
- AutoClip 前端(localhost:3000)显示的是完整话题段
- 最终切片(15-30秒)需要用 Whisper SRT 找精准高潮时间点,再用 FFmpeg 切
- 不要期望在 AutoClip 前端直接拿到短切片
从 AutoClip 片段切精剪(15-30秒)
1. 获取片段时间范围
curl -s "http://localhost:8000/api/projects/{PROJECT_ID}" | python3 -c "
import json, sys
data = json.load(sys.stdin)
clips = sorted(data['clips'], key=lambda x: x['final_score'], reverse=True)
for c in clips[:10]:
print(c['start_time'], c['end_time'], c['final_score'])
"2. 从 SRT 找爆点关键词时间戳
# 搜索138/350克/上车/扣1/埃及棉/万针/不变形等关键词
grep -n "138\|350\|扣一\|上车\|万针\|不变形\|压胶\|埃及\|销冠" \
/tmp/whisper_yoyo/yoyo_full.srt | head -303. 用 FFmpeg 切精剪片段(关键:terminal(background=true),不要用 &)
VIDEO="/path/to/video.mp4"
OUTPUT="/Users/yehya/Downloads/竞品视频/clips"
mkdir -p "$OUTPUT"
# 切30秒片段(从爆点时间前2秒开始)
ffmpeg -y -ss 00:07:19 -i "$VIDEO" -t 30 \
-c:v libx264 -crf 23 -preset fast \
-c:a aac -b:a 128k \
-movflags +faststart \
"$OUTPUT/clip_138价格揭示.mp4" 2>&1 | tail -2端口对照表:
- AutoClip 后端:
localhost:8000 - AutoClip 前端:
localhost:3000 - 前端路由:需通过
/api/...调用后端
已知问题 & 解决方案
| 问题 | 原因 | 解决 |
TypeError: unsupported operand for | | Python 3.9 不支持类型联合语法 | 用 Python 3.11 |
ModuleNotFoundError: dashscope | 依赖未安装 | pip install dashscope==1.23.5 |
Directory 'uploads' does not exist | 首次运行缺少目录 | mkdir -p uploads outputs logs |
vite: command not found | npm 依赖未安装 | npm install |
| Whisper 转录超时 | base 模型太慢 | 用 tiny 模型测试 |
| AutoClip 片段太长 | 正常,AI识别的是话题段 | 用 SRT+FFmpeg 再切精剪 |
| FFmpeg 背景编码失败 | 禁止在 foreground 命令里用 & | 用 terminal(background=true) |
| 竖屏视频素材方向 | 直播是 9:16 竖屏(1080×1920) | 输出保持原分辨率 |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

