video-maker is an open-source Codex skill for producing Bilibili tech and science explainer videos. Version 2 is the lean workflow: fewer source contracts, stricter production gates, local Qwen TTS, Remotion assembly, real visual QA, and a publish handoff.
- Content-first production: the topic, evidence, audience, and cognitive goal drive structure, visuals, voice, pacing, and cover.
- Minimal contracts:
content_contract.json,evidence_map.json,script.json,visual_plan.json,segments.json, andqa_report.json. - No fixed house style: no default palette, persona, creator benchmark, cover template, scene skeleton, or duration target.
- Publish gates: evidence, semantic motion, real rendered pixels, full audio review, custom cover, metadata, and submission result must be checked before release.
SKILL.md: the Codex skill entrypoint and workflow.agents/openai.yaml: plugin-facing metadata.references/schema.md: minimal v2 artifact schemas.references/acceptance-rubric.md: final review checklist and failure routing.scripts/bootstrap_project.py: creates a new v2 project and copies runtime helpers.scripts/quick_check.py: validates project contracts and publish gates.scripts/render_all.ps1: runs checks, props preparation, optional TTS, and Remotion render.scripts/doctor.py: checks local Qwen, ffmpeg, node, and npm availability.
git clone https://github.com/BraveCowNoFear/video-maker.git
cd video-maker
python .\scripts\bootstrap_project.py `
--topic "Why transformers scale differently" `
--slug "transformer-scaling"
python .\scripts\doctor.pyBy default, projects are created under:
%USERPROFILE%\Desktop\Video Maker\bilibili projects
Override locations with:
$env:VIDEO_MAKER_PROJECTS_ROOT = "D:\Bilibili Projects"
$env:VIDEO_MAKER_QWEN_ROOT = "D:\TTS\qwen3-tts-1.7b"
$env:VIDEO_MAKER_PYTHON = "C:\Path\To\python.exe"The main agent acts as coordinator. Full production typically uses six roles:
content-strategistscript-writervisual-architectvisual-qa-fixerproduction-engineeracceptance-reviewer
See SKILL.md for the operating rules.
MIT