ContentEngineAI：开源电商短视频生成流水线，从抓取到发布全自动

License: MIT | Status: Pre-Production

ContentEngineAI is an AI-powered pipeline for generating short, vertical (9:16) promotional videos for e-commerce products. It automates the complete workflow from scraping product data to delivering final videos with AI-generated scripts, voiceovers, and content-aware subtitles.

Key Features

• End-to-End Automation: Complete video production from scraping to publishing
• Social Media Publishing: Auto-scheduling with slot detection + cleanup after publish
• Link-in-Bio Integration: Auto-add affiliate links to bio page after publishing (Lnk.Bio)
• Affiliate Disclosure: Opt-in rendering of an affiliate program's literal identification phrase in the caption body, configurable for non-Amazon programs (off by default; the #ad disclosure is separate and applied to any render carrying a material connection)
• Published Products Registry: Track all published products in JSON/CSV format
• Batch Processing: Process hundreds of products with unified scrape + produce + publish pipeline
• Platform-Specific Metadata: AI-generated titles, captions, hashtags for YouTube, TikTok, Instagram
• Content-Aware Subtitles: Dynamic positioning with CTA-synchronized dual-line support
• Premium Audio: Google Chirp 3 HD and Gemini TTS voices with Whisper STT synchronization, channel-wide voice pinning

Quick Start

ContentEngineAI runs from a checkout, not from a PyPI install: poetry install provides its dependencies and everything runs as python -m src.<module>. It is deliberately not published as a package.

# 1. Setup (requires Python 3.12+, FFmpeg, Poetry)
git clone https://github.com/stkzlv/ContentEngineAI.git && cd ContentEngineAI
poetry install && poetry run playwright install
cp .env.example .env  # Configure API keys

# 1b. Animated captions (optional, and on by default in the bundled config)
poetry install --with pycaps && poetry run playwright install chromium

# 2. Generate a video
poetry run python -m src.scraper.amazon.scraper --product-ids B0BTYCRJSS --debug
poetry run python -m src.video.producer outputs/B0BTYCRJSS/data.json slideshow_images1 --debug

# Or render a topic, with no scraper run and no product directory.
# The same flags work on the global batch, which also publishes it.
# This profile has no product photography, so it writes the script first and
# then searches stock footage on phrases taken from the narration.
poetry run python -m src.video.producer slideshow_stock \
  --topic "Why your wifi keeps dropping" \
  --topic-description "Router placement, channel congestion, 2.4 vs 5GHz."

# 3. Batch pipeline (scrape + produce + publish)
poetry run python -m src.pipeline.global_batch \
  --keywords "wireless earbuds" \
  --profile slideshow_images1 \
  --platforms youtube tiktok \
  --debug

See Installation for complete setup instructions.

Two things the quickstart assumes:

• Animated captions need the optional pycaps group. The bundled config sets subtitle_engine: pycaps, so step 1b is what makes the shipped default work. Without it a run falls back to FFmpeg captions and warns once; that is a supported configuration, just not the one the config describes. The renderer runs without a display as of 2026-09-02; if its per-word screenshots time out, wrap the producer in xvfb-run -a. See Subtitles and Troubleshooting.
• Publishing needs an account before --platforms does anything. Set LATE_API_KEY in .env and connect YouTube, TikTok or Instagram in the scheduler first. Add --skip-publish to run the batch without it. See Publishing.

Documentation

Developer Documentation

License

MIT License - see LICENSE for details