AI 视频生成器 — 短内容流水线
概述
使用 AI 自动化创建短视频(TikTok、YouTube Shorts、Instagram Reels)的每个步骤:话题研究、脚本撰写、文字转语音旁白、素材匹配、字幕生成和最终合成。灵感来自 MoneyPrinterTurbo(53k+ 星)。
说明
步骤 1:设置环境
pip install anthropic openai requests moviepy pydub whisperx srt
sudo apt install ffmpeg # Linux — 或:brew install ffmpeg (macOS)
需要的 API 密钥: Anthropic 或 OpenAI(脚本)、ElevenLabs 或 OpenAI TTS(语音)、Pexels(免费素材)。
步骤 2:AI 脚本撰写
import anthropic
def generate_script(topic, duration_seconds=45):
"""为短内容生成优化的视频脚本。"""
client = anthropic.Anthropic()
prompt = f"""写一段 {duration_seconds} 秒的视频脚本,主题:{topic}
格式:
钩子(前 3 秒):令人震惊的陈述或问题,让观众停止滑动
主体(主要内容):3-5 个有力的事实或要点,每个 1-2 句话
行动号召(最后 5 秒):关注、点赞、评论
规则:
- 口语化,不使用复杂句子
- 每句话单独一行
- 约 {duration_seconds * 2.5:.0f} 字({duration_seconds}s,语速 150wpm)
- 使用强力词汇:秘密、震惊、没人告诉你、实际上
- 无表情符号或话题标签——这是旁白脚本
"""
response = client.messages.create(
model="claude-sonnet-4-20250514", max_tokens=500,
messages=[{"role": "user", "content": prompt}]
)
return response.content[0].text
步骤 3:文字转语音旁白
import requests, os
def generate_voice_elevenlabs(text, output_path='narration.mp3'):
"""使用 ElevenLabs 生成旁白。"""
url = "https://api.elevenlabs.io/v1/text-to-speech/21m00Tcm4TlvDq8ikWAM"
headers = {"xi-api-key": os.environ["ELEVENLABS_API_KEY"], "Content-Type": "application/json"}
data = {"text": text, "model_id": "eleven_turbo_v2_5",
"voice_settings": {"stability": 0.5, "similarity_boost": 0.75}}
response = requests.post(url, json=data, headers=headers)
with open(output_path, 'wb') as f:
f.write(response.content)
return output_path
def generate_voice_openai(text, output_path='narration.mp3'):
"""使用 OpenAI TTS 生成旁白(更便宜的选择)。"""
from openai import OpenAI
client = OpenAI()
response = client.audio.speech.create(model="tts-1-hd", voice="onyx", input=text)
response.stream_to_file(output_path)
return output_path
步骤 4:素材选择
def search_pexels_videos(query, count=5):
"""在 Pexels 搜索纵向方向的免版税视频片段。"""
url = "https://api.pexels.com/videos/search"
headers = {"Authorization": os.environ["PEXELS_API_KEY"]}
params = {"query": query, "per_page": count, "orientation": "portrait", "size": "medium"}
response = requests.get(url, headers=headers, params=params)
videos = response.json().get('videos', [])
results = []
for v in videos:
files = sorted(v['video_files'], key=lambda x: x.get('height', 0), reverse=True)
hd = next((f for f in files if f.get('height', 0) >= 720), files[0])
results.append({'id': v['id'], 'url': hd['link'], 'duration': v['duration']})
return results
步骤 5:字幕生成
def generate_subtitles(audio_path, output_srt='subtitles.srt'):
"""使用 WhisperX 生成词级字幕。"""
import whisperx, srt
from datetime import timedelta
model = whisperx.load_model("base", device="cpu")
audio = whisperx.load_audio(audio_path)
result = model.transcribe(audio)
align_model, metadata = whisperx.load_align_model(language_code="en")
aligned = whisperx.align(result["segments"], align_model, metadata, audio)
subs = []
words = [w for seg in aligned["segments"] for w in seg.get("words", [])]
for i in range(0, len(words), 4):
group = words[i:i + 4]
if not group: continue
start = timedelta(seconds=group[0].get('start', 0))
end = timedelta(seconds=group[-1].get('end', 0))
text = ' '.join(w['word'] for w in group)
subs.append(srt.Subtitle(index=len(subs)+1, start=start, end=end, content=text))
with open(output_srt, 'w') as f:
f.write(srt.compose(subs))
return output_srt
步骤 6:使用 FFmpeg 合成视频
import subprocess
def assemble_video(clips, narration, subtitles, output='final.mp4'):
"""合成最终视频:拼接片段、添加旁白和字幕。"""
concat_list = 'concat_list.txt'
with open(concat_list, 'w') as f:
for clip in clips:
f.write(f"file '{clip}'\n")
subprocess.run([
'ffmpeg', '-y', '-f', 'concat', '-safe', '0', '-i', concat_list,
'-vf', 'scale=1080:1920:force_original_aspect_ratio=increase,crop=1080:1920',
'-c:v', 'libx264', '-preset', 'fast', '-an', 'temp_video.mp4'
], check=True)
subtitle_filter = (f"subtitles={subtitles}:force_style='"
"FontName=Arial,FontSize=18,PrimaryColour=&H00FFFFFF,"
"OutlineColour=&H00000000,Outline=2,Bold=1,Alignment=2'")
subprocess.run([
'ffmpeg', '-y', '-i', 'temp_video.mp4', '-i', narration,
'-vf', subtitle_filter, '-c:v', 'libx264', '-c:a', 'aac',
'-shortest', output
], check=True)
return output
步骤 7:完整流水线
def generate_video(topic, output_dir='./output'):
"""完整流水线:主题 -> 完成视频。"""
import os
os.makedirs(output_dir, exist_ok=True)
script = generate_script(topic)
narration = generate_voice_elevenlabs(script, f'{output_dir}/narration.mp3')
keywords = topic.split()[:3]
videos = search_pexels_videos(' '.join(keywords), count=3)
clips = []
for i, v in enumerate(videos):
path = f'{output_dir}/clip_{i}.mp4'
requests.get(v['url'], stream=True) # 下载片段
clips.append(path)
subs = generate_subtitles(narration, f'{output_dir}/subs.srt')
return assemble_video(clips, narration, subs, f'{output_dir}/final.mp4')
示例
示例 1:批量生成科技事实短视频
一位创作者一次性为 TikTok 生成 5 个科技主题短视频:
topics = [
"没人谈论的 AI 工具",
"感觉像非法免费使用的应用",
"让你大开眼界的网站",
"每个学生都需要的免费 AI 工具",
"改变我生活的 50 美元以下科技小工具"
]
for topic in topics:
output = generate_video(topic, output_dir=f'./output/{topic[:30]}')
print(f"视频已就绪:{output}")
# 每个视频:约 45 秒,纵向 1080x1920,带字幕和旁白
# 总成本:约 $0.25(5 x $0.05 每个视频)使用 ElevenLabs + Claude Sonnet
示例 2:每日财经 Shorts 用于 YouTube
一个财经频道自动化每日 Shorts 上传,使用热门财经话题:
import schedule
def daily_finance_video():
topic = "三个真正有效的 2025 年被动收入点子"
script = generate_script(topic, duration_seconds=55)
# 脚本输出:
# 钩子:"你每天都在亏钱,因为你不知道这些。"
# 主体:1. 按需印刷商店($500-2k/月)
# 2. AI 生成内容授权($300-1k/月)
# 3. 股息 ETF 堆叠($200-800/月被动收入)
# 行动号召:"关注我,获取没人告诉你的理财技巧。"
narration = generate_voice_openai(script, './daily/narration.mp3')
videos = search_pexels_videos("money finance investing", count=4)
# 下载 4 个钱币/图表/生活方式纵向片段
# 用粗体白色字幕合成,输出 55 秒 Short
final = assemble_video(['./daily/clip_0.mp4', './daily/clip_1.mp4'],
narration, './daily/subs.srt', './daily/final.mp4')
schedule.every().day.at("08:00").do(daily_finance_video)
指导方针
- 尊重版权 — 仅使用免版税素材(Pexels、Pixabay)或你自己的内容
- 公开 AI 使用 — YouTube 和 TikTok 要求公开 AI 生成内容
- 发布前检查 — 始终观看最终视频;AI 脚本可能包含不准确信息
- 优化前 3 秒 — 钩子决定观众是停留还是滑动
- 测试多种语音 — ElevenLabs 提供数十种语音;找到适合你领域的声音
- 监控表现 — 跟踪观看次数、留存率和点击率,以迭代内容风格
参考
- MoneyPrinterTurbo — 原始灵感(53k 星)
- Pexels API — 免费素材视频
- ElevenLabs — 逼真文字转语音
- WhisperX — 词级字幕对齐