video-autopilot-kit: An Automated Video Pipeline for YouTube Shorts and Long-Form Content

Most short-form workflows still involve the same manual loop: cut the clip, add captions, check the pacing, export, repeat for every platform. video-autopilot-kit is an open-source, automated video pipeline for YouTube Shorts, long-form videos, and interview shows that replaces that loop with a set of Python scripts and, optionally, CapCut automation. It is built by developer Hao0321 and hosted on GitHub at github.com/Hao0321/video-autopilot-kit.

As of today, the repository has crossed 1,700 stars (1,763 at last check) and 289 forks, up from the 1,531 stars logged when it first appeared in earlier scans. It carries an MIT license, so commercial use, modification, and redistribution are all permitted with attribution.

What It Actually Does

The kit is not a single script that spits out a finished video. It is three separate production lines that share the same underlying pattern: take a set of inputs, run them through quality gates, and execute a one-command build.

• Long-form teaching videos: ffmpeg-driven effects (Ken Burns pans, bloom, light sweeps), word-level captions, and cleanup for raw screen recordings.
• Vertical Shorts: 9:16 normalization, multi-color subtitles, background-music peak detection, and loudness normalization, with length gates tuned per platform.
• Interview shows: guest-data validation plus a seven-item pre-production package (invitations, host scripts, prep sheets, release materials) generated from a template.

All three run on quality gates rather than raw output. Instead of just rendering and hoping it looks right, the pipeline checks for things like flicker, audio sync drift, and dead space before calling a video finished. That is the actual value: it is less "AI makes your video" and more "AI enforces the same checklist a good editor already runs by hand, every time, without you doing it manually."

CapCut Automation, or Skip CapCut Entirely

There are two ways to run this, and the choice matters for who it fits.

Path 1 (programmatic) works on Windows, Mac, and Linux with just Python 3.9+ and ffmpeg/ffprobe on PATH. No CapCut required. This is the cross-platform, fully scriptable option, and the one most creators will actually use day to day.

Path 2 (CapCut-assisted) is Windows-first and version-sensitive. It edits CapCut's own draft JSON files directly to pull in CapCut's built-in effects, cloud templates, and text overlays, and it leans on an AI assistant with computer-use capability to drive the CapCut GUI. This path adds real fragility: CapCut updates its app regularly, and the repo maintains a compatibility matrix specifically because draft-JSON edits can break across versions. If a creator wants a pure command-line pipeline with no GUI dependency, Path 1 is the one to reach for. Path 2 is for people specifically chasing CapCut's native look with less manual clicking.

What "Fill-In-Your-Own-Data" Actually Requires

This is the part worth being direct about, since the tool describes itself that way and the setup is real, not cosmetic.

The repo ships with zero private data or channel presets. Everything that shapes the output comes from a SETUP.md questionnaire covering channel niche, voice and vocabulary, KPIs, brand guidelines, and audience segmentation. That questionnaire generates local profile files (kept out of the repo via .gitignore) that the scripts read from. A config.example.py has to be copied to config.py and pointed at local source-footage and export paths before anything runs.

More notably, the vocabulary whitelists used for script quality gates ship intentionally empty. The project deliberately avoids preloading someone else's audience language, on the reasoning that borrowing another creator's phrasing filters would just clone their voice. Platform thresholds, like Shorts length and cut speed, come with example values that are meant to be recalibrated against a creator's own top three to five performing videos, not used as-is.

In practice, this means budget setup time before the automation pays off. This is not a plug-in-an-API-key tool. It is closer to a framework that needs a creator's own inputs to become useful, similar in spirit to filling out a detailed production bible before a pipeline can run on autopilot.

Who This Fits

This is built for creators and small content teams already producing short-form video at volume who want a repeatable system rather than a one-off edit. If the output is one video every couple of weeks, the setup overhead (questionnaire, config, calibration against real footage) will outweigh the time saved. It earns its keep once someone is publishing multiple Shorts a week and repeating the same manual QA checklist each time.

It is also not for anyone hoping to generate a full video from a text prompt with no footage. The tool assembles, captions, normalizes, and gates content; it does not create source footage or write the script. Someone still has to shoot or screen-record the raw material and supply the script text, or let the pipeline caption from reviewing video frames directly.

Getting Started

The basic path: run through SETUP.md, copy config.example.py to config.py and set local paths, pick Path 1 or Path 2, then run the relevant script, for example python src/shorts_autopilot.py scan for Shorts, followed by a review step and build. A system_health.py --quick check validates the install before a first real run.

Full setup instructions, the compatibility matrix, and the source code are at github.com/Hao0321/video-autopilot-kit.