A source toolkit for AI-assisted video editing: an editing handbook, operating instructions, validated production contracts, Python engines and finishing tools. It is for editors and developers who want an inspectable workflow with explicit human decisions.
Status: v0.1.0-rc.1 is published; this source proposes v0.2.0-rc.1. The expanded candidate has 209 core checks and separate style, graphics and native 3D browser profiles. Final fresh-candidate verification and publication review are required before the proposed release. See the verification scope and capability matrix. Real productions need job-specific evidence.
- Follow installation using a separate Python environment and external workspace.
- Complete getting started: diagnostics and synthetic contract validation, with no paid API or model download.
- Run the offline synthetic walkthrough to inspect timing, draft captions, encoded QA and gate refusals.
- Read architecture, then the production workflow before using your own footage.
- Choose a specialist in the skill index and try the generated-input examples.
After installation and local configuration, run these from the toolkit root:
$ToolkitPython = '../editing-pack-env/Scripts/python.exe'
& $ToolkitPython -B scripts/run_checks.py --config config/toolkit.local.json
& $ToolkitPython -B 'Editing_OS/Video Editing/Contracts/validate.py' production-context 'Editing_OS/Video Editing/Contracts/examples/production-context.v1.example.json'The first command runs core checks using synthetic media in your external workspace. The second verifies the example's schema and semantic constraints; it does not authorize editing. See getting started for expected results.
| Layer | Purpose |
|---|---|
| Handbook | Editorial craft and reasoning; empirical claims have scope limits |
| Operating instructions | Task entry, taste, routing, scene planning and quality requirements |
| Contracts | Context, intent, taste selection, approvals and evidence formats |
| Owned engines | Editorial timing, captions, graphics, orchestration and encoded QA |
| Workbench | Diagnostics, frame-grid drafts and conditional editable handoffs |
| Hosted image utility | Structured image requests with explicit paid execution controls |
The source checkout contains no user footage, job transcripts, real approvals, models, fonts, installed environments or third-party specialist clones. Keep jobs outside it. A .gitignore is an additional convenience, not a privacy boundary.
A real production job needs genuine forced alignment, a reviewed and hash-bound EDL, post-cut retimed words, explicit treatment selection, encoded-artifact evidence and exact-hash human review. A preview or passing test cannot supply those decisions. Workflow and review explains the sequence and the distinction between a review MP4, an editable handoff and publication readiness.
- Installation and configuration
- Getting started and troubleshooting
- Architecture, capabilities and optional tools
- Privacy and job workspaces
- Contributing, maintenance, changelog and security reporting
Owned code, tests, schemas and operational instructions are MIT. The handbook and explanatory documentation are CC BY 4.0, with MIT code snippets. See license scope, MIT, CC BY 4.0 and third-party notices. External tools, models and services retain their own terms.