autovideo

AutoVideo is a Python project that automates the production of short-form social media videos, targeting creators who want to batch-produce content in the style of popular accounts rather than make each video from scratch. The README is written in Chinese, but the concept translates directly: you give the system a handful of example videos from a viral creator, it reverse-engineers what makes that creator's style work (camera angles, pacing, text placement, color choices), and then packages those findings into a reusable recipe that can generate new videos automatically.

The author describes it as a "factory for making factories." Instead of a generic video generator that produces one-size-fits-all output, AutoVideo produces a pipeline tailored to a specific content category. Once the pipeline is locked in, creating new videos in that style means filling in a spreadsheet and clicking a button, with no manual filming or editing required.

The repo is structured in three layers. The first layer (A) handles measurement: it pulls raw data from reference videos, including timing, subtitles, and layout, without making any editorial judgment. The second layer (B) is an iterative loop where a human and AI collaborate to draft a recipe, generate sample output, compare it to the target, and refine until the results are close enough. The third layer (D) takes the finalized recipe and scaffolds a runnable pipeline project for that content category. A human review gate sits between the machine scoring and the final approval decision.

The repository is a showcase and learning reference, not the full production system. It publishes the architecture design, method documentation, and code scaffolding so other developers can study or adapt the approach. Three content categories are already demonstrated: AI book recommendation clips, dialogue-style educational content, and tech news highlights. The complete operational product, including the editing recipes, scoring engine, and batch production tools, lives at a separate website.

If you want to understand how the pipeline works or adapt the structure for your own content type, this repo gives you the blueprint. If you want to use the finished product directly, the author points to the external site.