Jianying Headless Lets Coding Agents Build Editable Video Drafts From JSON
A reverse-engineered toolkit lets AI agents drive CapCut's Chinese sibling Jianying entirely from the command line, producing native editable drafts and MP4 exports.
Takeaways
- jianying-headless drives ByteDance's Jianying editor from Python by generating native draft files instead of automating the GUI
- Repo hit 700+ stars in days after the author reversed plans to sell it for around 888 RMB
- Supports multi-track, PiP, subtitles, volume, local BGM; exports MP4 via Jianying's own engine in an isolated process
- Pinned to Apple Silicon macOS 26.0+ and Jianying Pro 11.4.2, with hash-verified builds and no auto-downgrades
- Drafts remain human-editable in the GUI, enabling agent-first, human-in-the-loop video workflows
- License is personal and non-commercial; see the repo for scope and third-party notices
Jianying Headless Turns JSON Editing Plans Into Native Video Drafts
jianying-headless is a source-available Python controller that generates Jianying project files and invokes the desktop app’s local rendering engine. Developers can use command-line scripts, CI jobs, or coding agents such as Codex to assemble editable drafts and export MP4 files.
Jianying is ByteDance’s Chinese desktop counterpart to CapCut. Its draft files store the timeline, clips, captions, audio tracks, effects, and related project state. Writing those files directly gives the automation a structured interface to the editor while preserving a project that can later be opened in Jianying for manual review.
| Component | Current behavior |
|---|---|
| Interface | Python command-line scripts |
| Input | JSON editing plan and local media |
| Output | Editable Jianying draft and optional render.mp4 |
| Editing scope | Segments, speed, volume, multiple tracks, picture-in-picture, captions, titles, background music, and sound effects |
| License | Personal learning and non-commercial use; commercial use requires written authorization |
The draft becomes the interface
The controller builds a native draft from an editing plan, modifies existing multi-track projects inside isolated copies, validates referenced resources, and starts the local exporter only after an explicit command. Keeping edits in Jianying’s project format also lets an editor inspect cuts, repair captions, rebalance audio, or continue working in the desktop application.
The repository separates draft generation from the proprietary bridge used during export:
engine/constructs drafts, creates isolated working copies, validates resources, and coordinates native export.bridge/provides file and pipe adapters for Jianying’s installed libraries.
Reporting is indexed from AlphaSignal. Rights remain with the original publisher and cited sources.
Read original report
