Turn Long Videos Into Social Clips Without Opening an Editor
You've got hours of long-form content sitting in a folder, and somewhere in there are a few moments that could absolutely kill on social media. But hunting through footage, cutting highlights, and adding subtitles is a slog you keep putting off. What if you could just drop a video file into a directory, run one command, and walk away while an AI crew does the editing for you? That's exactly what viral-clips-crew is built to do.
This is a CrewAI-powered video assistant that watches and listens to your long-form content, picks out the most striking segments, and spits out polished clips ready for publication. It's the brainchild of Alex Fazio, and honestly, it's the kind of tool that makes you wonder why you've been doing this manually.
What It Does
At its core, viral-clips-crew is a Python application that orchestrates a crew of AI agents to handle the entire video repurposing pipeline. You drag a video file into the input_files directory, run poetry run python app.py, and the system takes over from start to finish.
The architecture leans on CrewAI to coordinate the workflow, with both OpenAI and Google Gemini APIs doing the heavy lifting. The project uses Poetry for dependency management, which keeps the setup clean and reproducible. Everything you need is listed in the pyproject.toml file, so there's no hunting around for missing packages.
The final output lands in the subtitler_output directory, which is exactly what you'd expect: finished clips with subtitles baked in, ready to drop onto whatever social platform you're targeting.
One important detail: Gemini can handle videos up to an hour long, but if you're using the OpenAI API, you'll need to keep clips under 15 minutes. That's a hard constraint based on the current LLM context windows, so plan your input files accordingly.
Why It's Cool
The obvious appeal here is the time savings, but there's more to this project than just automation.
It treats content repurposing as a complete workflow, not a single step. You're not just getting transcription or just getting clip suggestions. The system handles the whole journey from raw footage to finished, subtitled clips. That's a genuinely useful abstraction for anyone who creates content regularly.
It's built on CrewAI, which means the architecture is modular by design. You can see how AI agents are being used to break down a complex task into discrete roles. If you're interested in multi-agent systems, this is a practical, real-world example of how to apply that pattern to something creative rather than just another chatbot wrapper.
The setup is refreshingly honest about its limitations. The README straight-up tells you this was "99% vibe coded as a fun Saturday hack" and that the author isn't planning to support it. There's no false promise of production-readiness, no enterprise sales pitch. It's a working prototype that's meant to inspire you to take it further.
It solves a real bottleneck for creators. The gap between having long-form content and getting short-form clips out the door is where most people lose momentum. This tool removes the friction of the editing phase entirely. You're not learning a video editor, you're not scrubbing through timelines. You're just dropping a file and checking back later.
How to Try It
Getting started is straightforward, assuming you have Python and API keys ready.
First, clone the repository:
git clone https://github.com/alexfazio/viral-clips-crew.git
Then install Poetry to manage dependencies:
pip install poetry
Install the required packages:
poetry install
Update Pydantic, because the project explicitly calls for it:
poetry update pydantic
Set up your environment file with both API keys:
echo -e "OPENAI_API_KEY=<your-api-key>
GEMINI_API_KEY=<your-api-key>" > .env
Now drop your video file into the input_files directory and run:
poetry run python app.py
The process kicks off automatically, and your finished clips will be waiting in the subtitler_output directory when it's done.
If you hit a TypeError: 'NoneType' object is not iterable error, the README points to two likely culprits: incorrect API keys in your .env file, or insufficient credits on your OpenAI and Google Cloud accounts. Worth checking those before diving deeper.
One thing to keep in mind: this project hardcodes API keys in the code and isn't meant for production use. For a weekend hack, that's fine. Just be mindful about where you run it.
Final Thoughts
viral-clips-crew is a clever, practical demonstration of what happens when you point AI agents at a tedious creative task. It's not a polished product, and the author doesn't pretend otherwise. But it is a working example of how CrewAI can be used to build something genuinely useful, and it's a great starting point if you want to experiment with AI-driven video workflows.
If you're a content creator who's been putting off repurposing your long-form material, this could save you real hours. If you're a developer curious about multi-agent systems, it's a compact, readable example to learn from. Either way, it's worth a look, and the MIT license means you can take it and make it your own.
Follow @githubprojects for more developer tools and open source projects.