jianying-headless turns a JSON plan into an editable Jianying Pro draft, 2,963 stars in 19 days
jianying-headless lets a coding agent write a JSON plan and turn it into an editable Jianying Pro video draft on macOS, exporting through Jianying's own engine. The repository reached 2,963 stars 19 days after its first public commit on 2026-09-15.
Image: GitHub
Why it mattersA team using an AI agent for short-form video can now hand the editor an editable draft, so the final cut, timing and music stay changeable before export.
A video an AI tool generates is usually a finished render you cannot open and change in a native editor afterwards.
A Python project called jianying-headless, by GitHub user mcncarl, lets a coding agent write a JSON editing plan and turn it into a multi-track Jianying Pro draft that opens in the real editor, then exports to MP4 through Jianying's own engine. The repository reached 2,963 stars on 2026-10-04, 19 days after its first public commit on 2026-09-15. Those are GitHub's own counts.
The project is a Python 3.9 or newer command-line tool for macOS Apple Silicon, built around Jianying Pro 11.5.0, with an 11.4.2 compatibility path. An editing plan written as JSON becomes an editable draft: video segmentation, multi-track composition, speed changes, volume, picture-in-picture, subtitles and titles. The author lists local-font support, linear keyframes for position, scale, rotation, opacity and volume, and six static geometric masks. The native engine on the user's own machine then exports the verified draft as an H.264 AAC MP4. An existing Jianying project can be edited inside an isolated copy, leaving the original untouched.
The repository also ships a standalone Agent Skill at skills/yichen-jianying-edit/. The author documents it as the entry point a coding agent uses to drive the pipeline, with a plan format and example JSON for both basic clips and a Hypit-to-Jianying handoff.
The author published a case rebuilding a 50.23-second Instagram scrolling animation tutorial that originated in Hypit. The transferred project holds 23 tracks and 154 clips in total: 8 video and image tracks with 38 clips, 1 voice track with 7 clips, 14 editable text tracks with 109 clips, and 39 source material files. The export passed a 1,507 of 1,507 frame check. The author warns in the same page that the handoff moves the project structure, with specific fonts, per-word colour animation, and some crops and shadows preserved differently or lost. A full subjective audio and video sign-off has not been done.
The license covers personal learning and non-commercial use. Commercial use needs written permission from the author. The code sits outside the usual MIT or Apache-2.0 grants that developers often assume on GitHub. The tool depends on a Jianying Pro installation at a specific version with a matching native library hash, build and signing identity. An unknown or non-matching component makes the tool stop, and there is no override to relax the check. The Jianying engine, account data, project materials and effect assets are not distributed with the source.
A separate Windows path, contributed in PR number 1 by instantgoing, renders an MP4 through FFmpeg from a validated plan snapshot. The output is a finished MP4; opening the result in the Jianying editor to re-edit is a macOS-only path.
For a team producing short video at scale, this is a way to put an AI agent inside the editing workflow while a human editor still handles the final step. The agent writes a structured plan, the Jianying file opens on the editor's machine, and the editor keeps the ability to change the cut, timing or music before the final render.
Source
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.