AI Editor

Describe an edit once. Open it in Final Cut, Resolve, Premiere, Kdenlive, Shotcut, Blender or CapCut.

How it works · Editors · Features · AI agent · Quick start · API · Open in editor · Voiceover · Testing · FAQ · Credits

tl = Timeline("demo")
tl.add_clip("interview.mp4", dur=8, src=12, fade_in=0.5)
tl.add_clip("broll.mp4", dur=5, crossfade=1)
tl.add_audio("music.mp3", volume=0.4, fade_out=2)
tl.add_title("Opening scene", start=1, dur=3, fade=0.5)
export(tl, "fcp", "out")        # out/demo.fcpxml

How it works

flowchart LR
    A[Your footage<br/>+ what you want] --> B[You, a script,<br/>or an AI agent]
    B --> C[(Timeline<br/>plain JSON)]
    C --> D1[Final Cut]
    C --> D2[Resolve]
    C --> D3[Premiere]
    C --> D4[Kdenlive / Shotcut]
    C --> D5[Blender]
    C --> D6[CapCut]
    D1 & D2 & D3 & D4 & D5 & D6 --> E[Open, polish, render]

Your footage is never touched. The project file only says which part of which file plays where. You or your agent decide the cuts. This project writes the project file.

Editors

Editor	Output	Tested
Final Cut Pro	`.fcpxml`	🟡
DaVinci Resolve	`.fcpxml` + import script	🟡
Premiere Pro	`.xml` + `.srt`	🟡
Kdenlive / Shotcut	`.mlt`	🟢
Blender	`.py`	🟢
CapCut	draft folder	🟡

🟢 rendered and measured 🟡 file checked, not opened in the app. iMovie and Filmora can't import a timeline, so they're not supported.

Features

mindmap
  root((AI Editor))
    Video
      Trim and sequence
      Overlay tracks
      Still images
      Speed
      Opacity
      Fades
      Crossfade
      Zoom keyframes
    Audio
      Multiple lanes
      Volume
      Fades
      Voiceover
    Text
      Titles
      Colour, size, position
      SRT export
    Project
      Resolution and fps
      Save / load JSON
      CLI and Python

Every editor supports every feature. A test fails if one doesn't.

Feature	Final Cut	Resolve	Premiere	Kdenlive / Shotcut	Blender	CapCut
Trim, sequence, tracks	✓	✓	✓	✓	✓	✓
Volume, speed, opacity	✓	✓	✓	✓	✓	✓
Fades, crossfade	✓	✓	✓	✓	✓	✓ ¹
Titles	✓	✓	✓	✓	✓	✓
Zoom keyframes	✓	✓	✓	✓	✓	✓
Still images	✓	✓	✓	✓	✓	✓

¹ Pictures only. For audio, use separate tracks with fade_in/fade_out.

Not included: effects, filters, colour grading, other transitions (each editor has its own), font choice and text animation, and reading existing projects. Media is referenced by absolute path, so moving it means re-linking in the editor.

AI agent

Not a chat app, but any coding agent (Claude Code, Codex, Cursor) can drive it, because an edit is just a few lines of Python or a JSON file.

sequenceDiagram
    actor You
    participant Agent as AI agent
    participant AE as AI Editor
    participant Ed as Your editor
    You->>Agent: "30 s highlight, title, music, export for Premiere"
    Agent->>Agent: look at footage (ffprobe, frames, transcript)
    Agent->>AE: write Timeline
    AE->>Ed: project file
    You->>Ed: review, adjust, render

The agent can only cut on what it can see or hear. Give it a transcript or notes for best results.

Try saying

"Cut ~/footage/trip.mp4 to 30 seconds with a title, and export for Premiere." "Add song.mp3 quietly under the clips and dissolve between them." "Zoom in on each click at 3.2 s, 7.8 s and 12.5 s, then export for Resolve."

Not built yet: a ready-made agent instruction file, and helpers for transcription and scene detection.

Quick start

Needs: Python 3.10+, ffmpeg (brew install ffmpeg). The editor is only needed to open the result.

git clone https://github.com/zandy700/AI-Editor.git && cd AI-Editor
python3 -m venv .venv && source .venv/bin/activate
pip install -e .                     # core
pip install -e '.[narrate,capcut]'   # optional: voiceover, CapCut

Make two test clips, then export:

ffmpeg -f lavfi -i testsrc=duration=10:size=1280x720:rate=30 -f lavfi -i sine=duration=10 -shortest -pix_fmt yuv420p a.mp4
ffmpeg -f lavfi -i smptebars=duration=10:size=1280x720:rate=30 -f lavfi -i sine=frequency=330:duration=10 -shortest -pix_fmt yuv420p b.mp4

from editor import Timeline
from editor.export import export

tl = Timeline("demo")
tl.add_clip("a.mp4", dur=4, src=2, fade_in=0.5)
tl.add_clip("b.mp4", dur=4, crossfade=1)
tl.add_title("Hello", start=1, dur=2, fade=0.5)
export(tl, "fcp", "out")            # or premiere, resolve, kdenlive, shotcut, blender, capcut

Command line, from a saved timeline:

python -m editor export timeline.json fcp premiere -o out
python -m editor export timeline.json all -o out

API

Timeline(name="Edit", width=1920, height=1080, fps=30). All times are in seconds.

 track 1 (overlay)   ░░░░░ watermark ░░░░░░░░░░░░░░░░░░░░░░
 track 0 (main)      ▓▓▓ a.mp4 ▓▓▓▓▓▓╲╱▓▓▓▓ b.mp4 ▓▓▓▓▓▓▓▓▓
                                      ↑ crossfade overlap
 title                    [ Hello ]
 audio 0             ♪♪♪♪♪♪♪♪♪♪ music.mp3 ♪♪♪♪♪♪♪♪♪♪♪♪♪♪
 time  0s ───────────────────────────────────────────────▶

Method	Does
`add_clip(path, dur, src, start, track, **opts)`	Video or still image
`add_audio(path, dur, src, start, track, **opts)`	Audio on a lane
`add_title(text, start, dur, **opts)`	On-screen text
`save(path)` / `Timeline.load(path)`	JSON round trip
`srt()`	Titles as subtitles

Defaults that keep scripts short: start is the end of the track, so repeated add_clip calls chain; dur is the rest of the file (stills need dur); track=0 is the main video.

Clip option	Default	Meaning
`src`	0	Trim: in-point in the file
`volume`	1.0	Sound level
`speed`	1.0	2.0 = twice as fast. `dur` is the length on the timeline, so a 2× clip with `dur=2` uses 4 s of the file
`opacity`	1.0	Picture transparency
`fade_in` / `fade_out`	0	Seconds. Picture fades from/to black (transparent on overlays) and sound from/to silence
`crossfade`	0	Seconds. Overlaps the previous clip on its track and dissolves across the overlap. `start` is set for you (previous end − crossfade), so the timeline gets shorter. Main video track only
`zoom`	[]	`[(seconds_into_clip, scale), …]`, 1.0 = no zoom. Usually set by `apply_click_zoom`
`track`	0	Video: 0 = main, 1+ = overlays on top. Audio: lane number

Title option	Default	Meaning
`y`	-0.8	-1 bottom to 1 top (default is subtitle position)
`size`	60	Pixels at 1080p
`color`	`#ffffff`	Hex
`fade`	0	Fade in/out seconds

Click zoom for screen recordings:

from editor.zoom import apply_click_zoom
rec = tl.add_clip("screen.mp4")
apply_click_zoom(rec, [3.2, 7.8, 12.5], scale=1.5)   # times in seconds from clip start, or a JSON file of times

scale=1.5 zoom factor, ramp=0.25 seconds to zoom in/out, hold=0.6 seconds to stay zoomed.

JSON (write it from any language, or have an LLM produce it). tl.save("timeline.json") and Timeline.load(...) round-trip the whole edit. Use absolute paths. start is required on every clip in JSON.

{
  "name": "demo", "width": 1920, "height": 1080, "fps": 30,
  "clips": [
    {"path": "/abs/path/a.mp4", "start": 0, "dur": 4, "src": 2, "kind": "video", "track": 0,
     "volume": 1.0, "speed": 1.0, "opacity": 1.0, "fade_in": 0.5, "fade_out": 0.0,
     "crossfade": 0.0, "zoom": []},
    {"path": "/abs/path/music.mp3", "start": 0, "dur": 8, "src": 0, "kind": "audio", "track": 0,
     "volume": 0.5, "speed": 1.0, "opacity": 1.0, "fade_in": 1.0, "fade_out": 2.0,
     "crossfade": 0.0, "zoom": []}
  ],
  "titles": [
    {"text": "Hello", "start": 1, "dur": 2, "y": -0.8, "size": 60, "fade": 0.5, "color": "#ffffff"}
  ]
}

Open in your editor

Editor	Files	How
Final Cut	`Name.fcpxml`	File → Import → XML. A new event named after the timeline appears
Resolve	`Name.fcpxml`, `Name.resolve.py`	File → Import → Timeline, or run the script (below)
Premiere	`Name.premiere.xml`, `Name.srt`	File → Import. If you have titles, import the `.srt` and drag it on as captions, in case Premiere drops the XML titles
Shotcut	`Name.mlt`	File → Open File
Kdenlive	`Name.mlt`	File → Open. If it refuses, open in Shotcut or render with `melt`
Blender	`Name.blender.py`	See below
CapCut	folder `Name/`	Export into CapCut's drafts folder, restart CapCut

Resolve script, with Resolve running:

python3 out/Name.resolve.py                    # import the timeline
python3 out/Name.resolve.py --render ~/Movies  # import, then queue and run a render

Enable DaVinci Resolve → Preferences → System → General → External scripting using: Local. Some versions limit scripting to Studio. Importing the .fcpxml by hand always works.

Blender (5.x tested):

blender -b --python out/Name.blender.py                         # build only
blender -b --python out/Name.blender.py -- --render video.mp4   # build and render

macOS executable: /Applications/Blender.app/Contents/MacOS/Blender. Run without -b to edit by hand in the Video Editing workspace.

CapCut (pip install -e '.[capcut]'). Pass the full drafts path:

export(tl, "capcut", "/Users/you/Movies/CapCut/User Data/Projects/com.lveditor.draft")              # macOS
export(tl, "capcut", r"C:\Users\you\AppData\Local\CapCut\User Data\Projects\com.lveditor.draft")   # Windows

The location differs between versions, and newer versions may encrypt drafts and refuse them.

No editor at all: render the MLT file with melt (brew install mlt):

melt out/Name.mlt -consumer avformat:video.mp4 vcodec=libx264 acodec=aac

Voiceover

from editor.narrate import narrate
narrate(tl, "Welcome back. Today we cut a short film.")
narrate(tl, "这是一个中文句子。", start=12)    # Chinese picks the Chinese voice

Each sentence is spoken, placed on the timeline, and gets a subtitle of the same length.

Option	Default
`voice`	`en-US-AriaNeural`
`zh_voice`	`zh-CN-XiaoxiaoNeural`
`start`	after existing audio
`gap`	0.1 s between sentences
`out_dir`	temp folder. Pass one you keep, the exported project points at those mp3s

Needs pip install -e '.[narrate]' and internet. It uses edge-tts, which calls Microsoft's read-aloud service. That service is unofficial and could change or stop, so don't build anything critical on it.

Testing

Target	Checked by
🟢 Kdenlive / Shotcut	Rendered with `melt`; colours, timing, titles, audio levels measured
🟢 Blender	Rendered headless in Blender 5.2; same measurements
🟡 Final Cut, Resolve, Premiere, CapCut	Valid file, references resolve, times frame-aligned. Never opened in the real app

Not opened in the real apps means some details are unconfirmed: Final Cut's title template ID and timeMap, Premiere's effect IDs and text generator. Resolve may ignore speed changes, keyframed fades or title styling. If an export looks wrong, open an issue with the editor, version, screenshot and your timeline .json.

pip install -e '.[narrate,capcut]'
for t in tests/test_*.py; do python "$t"; done

editor/
  timeline.py   model + media probing
  narrate.py    voiceover + subtitles
  zoom.py       click zoom
  export/       fcpxml, resolve, premiere, mlt, blender, capcut
tests/          one file per exporter, plus features.py (shared full-feature timeline)

Add an editor: write editor/export/<name>.py with SUPPORTS = set(FEATURES) and export(timeline, out_dir), register it in editor/export/__init__.py, add a test.

FAQ

Can't see the project? Import the file as shown in Open in your editor. For CapCut, restart the app.
Footage missing? Paths are absolute. Re-link if you moved it.
Looks wrong? Open an issue (see Testing).

License

MIT

Credits

This project was inspired by jianying-editor-skill, an AI skill that builds video edits for JianYing from plain-language requests. Its idea of letting an AI agent write the timeline is what AI Editor carries over to other editors. No code was copied. Thank you to the people who built it:

_luoluoluo22
_{Project author / Maintainer}

_twodogegg
_{macOS compatibility / ffprobe fallback / unit tests}

_shaozheliu
_{Missing-media fix / self-contained assets}

_Maxinsomnia
_{macOS 5.9+ execution support}

The optional CapCut export uses pyJianYingDraft by GuanYixuan (Apache-2.0), installed as a separate dependency.

Name		Name	Last commit message	Last commit date
editor		editor
tests		tests
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
pyproject.toml		pyproject.toml

Navigation Menu

Repository navigation

About

Repository files navigation

AI Editor

How it works

Editors

Features

AI agent

Quick start

API

Open in your editor

Voiceover

Testing

FAQ

License

Credits

About

Releases

Packages

Contributors

Languages

Latest commit

History

Folders and files

Repository files navigation

AI Editor

How it works

Editors

Features

AI agent

Quick start

API

Open in your editor

Voiceover

Testing

FAQ

License

Credits

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages