MIT 许可的 agent skill，基于 Atlas Cloud 与本地 ffmpeg，把单一选题自动制成 Vox 风拼贴解说视频；可安装到任何 coding agent 的 skills 目录，包括 DeepSeek Harness 工作流。
DSH 适配
兼容
自动推断
安全审计
未审计
最后核验
2026-08-21
许可证
MIT
01它能帮你完成什么？
Turn a one-line topic into a finished Vox-style paper-collage explainer or ad video
A final MP4 with script, collage keyframes, motion, voice-over, music and captions, assembled at out/<project>/final.mp4 after two human approval gates
Creators and marketers using DeepSeek Harness or another coding agent who need editorial-style explainer/ad videos without hand-editing them
Re-style existing footage or photos into the collage look (A-roll / C-roll)
Talking-head videos re-styled frame-for-frame while keeping the real face, lip-sync and gestures; still photos cut out as photographic stickers anchored into generated posters, with optional voice-cloned narration
Video producers and brand teams that already have talking-head footage or product shots and want the Vox collage aesthetic
02如何接入 DeepSeek Harness？
前置条件
- A coding agent — Claude Code, Codex, or similar
- An Atlas Cloud API key
- ffmpeg + ffprobe (e.g. `brew install ffmpeg`)
- Python 3 with Pillow (`pip install pillow`) for caption/watermark overlays
安装步骤
- 01
Option A — clone the repo as a skill: `git clone https://github.com/Alisa0808/vox-director.git ~/.claude/skills/vox-director`
$ git clone https://github.com/Alisa0808/vox-director.git ~/.claude/skills/vox-director
- 02
Option B — download `vox-director.skill` and install it via your Claude skills UI
- 03
Set your Atlas Cloud API key: `export ATLASCLOUD_API_KEY="sk-..."`
验证接入成功
作者未说明
03DSH 适配与能力边界
Agent skill (SKILL.md workflow + scripts) installed into a coding agent's skills directory; the README targets any coding agent and does not mention DeepSeek Harness explicitly, though the repo is tagged dsh-plugin.
End-to-end video pipeline with human gates
A one-line topic (B-roll), a talking-head video (A-roll), or a still photo (C-roll)→beat map → style bake-off → collage keyframes → motion → voice + music → ffmpeg assembly → final.mp4
Sends prompts and media to the Atlas Cloud API (network calls to api.atlascloud.ai)Writes generated assets and final.mp4 under out/<project>/Three input modalities (B-roll / A-roll / C-roll)
a topic, a talking-head video, or a still photo→fully generated film; talking-head re-styled with face/lip-sync preserved; photo anchored as a photographic sticker with optional voice cloning
Voice cloning (bytedance/seed-audio-1.0) processes a real person's voice via Atlas CloudLocal ffmpeg assembly & overlays
generated clips, narration, music and captions→final.mp4 with concatenated clips, music ducked under VO, burned-in captions and watermark
Requires local ffmpeg/ffprobe and Python 3 + Pillow for caption/watermark overlays
04适合谁？何时不该用？
适合
- Creators and marketers using DeepSeek Harness or another coding agent who need editorial-style explainer/ad videos without hand-editing them
- Video producers and brand teams that already have talking-head footage or product shots and want the Vox collage aesthetic
不适合
- All image, video, voice and music generation runs through the Atlas Cloud API — you need an Atlas Cloud API key and network access, and model usage is billed against your Atlas Cloud account.
05兼容性、维护与安全提示
- All image, video, voice and music generation runs through the Atlas Cloud API — you need an Atlas Cloud API key and network access, and model usage is billed against your Atlas Cloud account.
- Model IDs on Atlas Cloud drift over time; the skill fetches the live model list before every run, and the model table is only verified on Atlas Cloud.
- Final assembly and caption/watermark overlays depend on locally installed ffmpeg + ffprobe and Python 3 with Pillow; the skill cannot produce the final MP4 without them.
MIT · actively maintained (last push 2026-08-11, no tagged releases yet)
06常见问题
Vox Director 如何接入 DeepSeek Harness？是原生还是 MCP？
都不是——它是一个 agent skill，不是 MCP 服务或原生运行时。README 表示任何能读取工作流并运行脚本的 coding agent 都可以使用：克隆到 skills 目录（Claude Code 会自动发现），或安装打包的 .skill 文件；其他 agent 读取 AGENTS.md → SKILL.md。README 并未专门提及 DeepSeek Harness。
安装前需要准备什么？
一个 coding agent（Claude Code、Codex 或类似工具）、Atlas Cloud API key、ffmpeg + ffprobe，以及安装了 Pillow 的 Python 3（用于字幕/水印叠加）。
我的数据和请求会去向哪里？
所有生成环节——关键帧、动画、配音、配乐——都会发送到 Atlas Cloud API（api.atlascloud.ai），每次运行前还会拉取实时模型列表；最终合成在本地用 ffmpeg 完成，输出到 out/<project>/final.mp4。
制作过程中我能保留多少控制权？
两个人工决策门：你需要确认节拍图（叙事结构），并从 3–4 种主题的风格比选中挑选视觉风格；之后的所有步骤全部自动完成。
07相关的 DSH 工作流
dsh-voice-ai-girlfriend
作者 beiyege-01
语音 AI 女友（Voice AI girlfriend for DeepSeek Harness）：Whisper 语音输入 + Qwen3-TTS 声音克隆 + 句子级流式朗读 + 数字人动画窗。插话/排队双模式，说话即打断。
dsh-plugin-tts
作者 1624318455
Edge TTS 语音朗读插件：朗读回复、自动朗读开关，免费无需 API key
dsh-ears
作者 wiziscool
DeepSeek Harness 语音输入插件，支持多后端打磨语音输入能力（ASR/Whisper）
dsh-speak
作者 alan2z
一种能让你的harness开口说话的方案/Make your AI harness speak — a verified voice-announcement solution
08数据与来源
usable by any coding agent (Claude Code, Codex, etc.). You give it a one-line topic; it gives you an `mp4`.
This is an **agent skill** — it works with any coding agent that can read a workflow and run scripts (Claude Code, Codex…
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-08-21。发现错误？提交更正。