Cordis dual-engine video workbench for DeepSeek Harness: War Room, beat edits, optional BYOK cloud media, local TTS. DSH bundle install includes FFmpeg/Remotion on desktop/web; standalone at `127.0.0.1:4399`.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-20
License
MIT
01What can it help you accomplish?
Produce AI-directed videos from DSH without leaving the desktop client
MP4 exports from a four-agent War Room (plan → edit → voice → render) plus step studio timeline tweaks
DSH users who want agent-assisted video creation with a visual workbench tied to their chat session
Remake a reference clip's pacing and style on a new topic
Reusable `MOTION.md` and brand kit after reference analysis, then a new rendered video
Creators who upload a reference video and want same-style beats, cuts, and look on different content
02How to install into DeepSeek Harness
Prerequisites
- Node.js `>= 20.0.0`
- FFmpeg and `ffprobe` on the system PATH (or paths set later in `.env`)
- Microsoft Edge or Chrome for headless HTML5 / Remotion rendering (per README hardware section)
Installation steps
- 01
Run `dsh plugin --profile desktop add github:Merziller/dsh-video-workbench` to install the self-contained bundle — README says Remotion and FFmpeg auto-install with the package without manual `FFMPEG_PATH` setup; on Windows it can reuse system Microsoft Edge for Remotion and HTML5 rendering
$ dsh plugin --profile desktop add github:Merziller/dsh-video-workbench
- 02
Alternatively use the DSH plugin market install button when the card appears — README says no manual dependency wiring is needed for that path
- 03
For source development: clone the repo, run `pnpm install`, then `node scripts/install-profile.mjs`, and fully quit and restart the DeepSeek Harness desktop client
- 04
Optional: copy `.env.example` to `.env` when you need custom FFmpeg paths, local IndexTTS, or Remotion template roots
Verify the integration
- Run `node scripts/self-test.mjs` for the README's 16-step end-to-end self-test (FFmpeg, Cordis tools, render lanes, and bundle checks)
03DSH integration and capability boundaries
Install as a Cordis DSH bundle, open the AI Video Workbench from the DSH sidebar, and sync LLM settings with `follow-dsh` or plugin-local config.
DSH War Room and step studio
Images, video clips, BGM, notes, and director prompts from the workbench UI→Orchestrated agent runs, previews, timeline edits, and packaged MP4
Serves the workbench UI via local REST under `/api/video-workbench/*`Spawns and supervises child processes for rendering and optional TTS servicesRemotion and HTML5 dual-engine render
Scene plans and TSX/HTML5 compositions→H.264 MP4 at chosen resolution and frame rate (720p–4K, 30/60fps per UI)
Invokes FFmpeg for encode, beat analysis, and keyframe extractionLaunches headless Chromium CDP and/or `@remotion/renderer` on the hostMulti-engine voice and beat sync
Scripts, BGM uploads, and TTS engine choice (local IndexTTS / GPT-SoVITS / Edge-TTS / beat-only MV)→Ducked narration mixes or beat-synced cuts without TTS
May start local IndexTTS HTTP services using paths from `.env`Reads and writes audio assets under the project workspaceLLM director via four API shapes
Chat prompts plus optional uploaded media context→Structured director plans, polished prompts, and coded scene updates
Calls configured LLM endpoints (OpenAI, Anthropic, Gemini, or Responses-compatible) — can follow DSH client keys via `follow-dsh`Optional BYOK cloud image, video, and music
User-supplied API keys and prompts in the cloud media center→Generated stills, clips, or music tracks merged into the project
Sends prompts and media job payloads to third-party HTTP APIs (Seedream, Seedance, Kling, Flux, Suno, MiniMax, ElevenLabs, or custom JSON endpoints)
04Who is it for? When not to use it?
Good for
- DSH users who want agent-assisted video creation with a visual workbench tied to their chat session
- Creators who upload a reference video and want same-style beats, cuts, and look on different content
Not for
- Optional BYOK cloud image, video, and music providers receive your prompts and generation requests over HTTPS when you enable them — content leaves the machine.
- README requires Node.js `>= 20.0.0`, system FFmpeg/ffprobe, and Edge or Chrome; local IndexTTS or 2K/4K 60fps work needs stronger GPU/RAM.
- API keys, FFmpeg paths, and TTS endpoints are loaded from an untracked `.env` file — you are responsible for what gets sent to LLM and cloud media APIs.
05Compatibility, maintenance and safety notes
- Optional BYOK cloud image, video, and music providers receive your prompts and generation requests over HTTPS when you enable them — content leaves the machine.
- README requires Node.js `>= 20.0.0`, system FFmpeg/ffprobe, and Edge or Chrome; local IndexTTS or 2K/4K 60fps work needs stronger GPU/RAM.
- API keys, FFmpeg paths, and TTS endpoints are loaded from an untracked `.env` file — you are responsible for what gets sent to LLM and cloud media APIs.
- README releases under MIT and requires keeping Merziller attribution and contact info on redistribution, forks, or repackaging.
MIT · last push 2026-11-19 · release v0.5.0
06Frequently asked questions
What is the recommended install path for DSH desktop?
README recommends `dsh plugin --profile desktop add github:Merziller/dsh-video-workbench`, which ships Remotion and FFmpeg installers so you skip manual template setup. The DSH market card offers the same silent install. After a source install with `node scripts/install-profile.mjs`, fully quit and restart DSH before opening the sidebar workbench.
Do LLM calls reuse my DSH API keys?
The workbench can follow DSH with `follow-dsh` so model endpoints and keys mirror the desktop client, or you can configure OpenAI, Anthropic, Gemini, or Responses-compatible endpoints inside the plugin. Polishing and coding steps hit those endpoints over the network.
What happens if I enable cloud image or music providers?
Cloud centers are optional BYOK. When turned on, prompts and job payloads go to vendors such as Seedream, Kling, Flux, Suno, or MiniMax, including custom HTTP JSON adapters. Disable them to keep generation entirely local except LLM calls.
Can I run it without local GPU TTS?
Yes. README documents beat-only MV mode without TTS, Windows native voice, Edge-TTS, and cloud speech templates. Local IndexTTS-2.5 or GPT-SoVITS needs the extra Python/GPU setup described in the TTS section and optional `node ./bin/dsh-video-cli.mjs install-tts`.
How do I confirm the bundle is healthy?
Run `node scripts/self-test.mjs` after install. It exercises FFmpeg probes, Cordis tool registration, render lanes, DSH sync slots, and bundle hygiene checks listed in the README verification section.
07Related DSH workflows
dsh-agent-teams
by nanmicoder
AgentTeams 多智能体团队协作插件:自然语言驱动建团队、拉成员、拆任务声明依赖、成员间消息直达,Web 右上角活动面板
dsh-blender-plugin
by sixtysevenlf
Adds tools for Blender scene editing, viewport capture, render profiling, headless work, and verification.
dsh-free-search
by dddmuc
DSH 免费搜索插件,执行联网检索。
dsh-bili-scroll
by c1overzz
Schedules Bilibili discovery, reads subtitles, writes daily reports, and optionally posts a comment tagging you.
08Data and sources
dsh plugin --profile desktop add github:Merziller/dsh-video-workbench
接入 DeepSeek Harness 桌面端后,可通过左侧栏入口切换进入工作台
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-20. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.
