An LLM assembly line for TikTok and YouTube Shorts

What it does

This Python framework automates short-form video production end-to-end. It generates scripts via OpenAI, synthesizes voiceovers in dozens of languages, sources background footage from Pexels and Bing, burns in captions, and renders the final cut. The package exposes three main engines: one for YouTube Shorts and TikToks, one for longer videos, and one that dubs existing content into other languages.

The interesting bit

The framework treats video editing as an LLM-readable task by expressing edits in a custom JSON-based “Editing Markup Language.” Instead of manually cutting timelines, you describe the assembly in structured text, and MoviePy executes it. That abstraction keeps the entire pipeline programmable and automatable.

Key highlights

• Three engines: ContentShortEngine, ContentVideoEngine, and ContentTranslationEngine for dubbing and translation
• Voice synthesis via both ElevenLabs and free Microsoft EdgeTTS, supporting 30+ languages
• Automatic asset sourcing from the web, Pexels, and Bing Images
• Persistent state managed through TinyDB
• Ships with a Gradio web UI and a ready-to-run Google Colab notebook

Caveats

• Explicitly labeled “experimental,” and the README admits more documentation is incoming
• Local deployment requires Docker; setup details are offloaded to a separate installation-notes.md
• Heavy reliance on external services: OpenAI for scripting, ElevenLabs/EdgeTTS for voice, and Pexels/Bing for assets

Verdict

Content creators and developers building automated short-form channels should look here. Anyone seeking fine-grained, human-in-the-loop video editing will likely find the LLM abstraction layer limiting.