Merziller/dsh-video-workbench

The workbench offers two interfaces in the DeepSeek Harness sidebar: a "Command Center" dashboard and step-by-step refinement. Four agents—planning director, material editor, voiceover and mixing, and motion graphics quality assurance and export—work in relay, allowing users to pause at any time and retry individual steps. The visuals rely on two engines: large language models generate Remotion TSX animations shot by shot (with automatic fixes via static and type checking), or HTML5 motion effects are used to render the final video directly. After uploading background music, FFmpeg is used to detect beats and climaxes for precise timing. The model interface is compatible with OpenAI, Anthropic, and Gemini protocols, and can also follow DSH settings.

The workbench offers two interfaces in the DeepSeek Harness sidebar: a "Command Center" dashboard and step-by-step refinement. Four agents—planning director, material editor, voiceover and mixing, and motion graphics quality assurance and export—work in relay, allowing users to pause at any time and retry individual steps. The visuals rely on two engines: large language models generate Remotion TSX animations shot by shot (with automatic fixes via static and type checking), or HTML5 motion effects are used to render the final video directly. After uploading background music, FFmpeg is used to detect beats and climaxes for precise timing. The model interface is compatible with OpenAI, Anthropic, and Gemini protocols, and can also follow DSH settings.

Why it matters A one-sentence request can be expanded into directorial instructions, with shots cut to the background music's beat. Voiceovers can connect to local IndexTTS or GPT-SoVITS, and the final video can be reviewed and modified shot by shot.

Stars 34

Reason

  • The repository 11 was created on 14, under the MIT license; the README specifies the prerequisites for each engine and the Windows 11 installation steps

Usage

  • Run dsh plugin --profile desktop add github:Merziller/dsh-video-workbench; for local voiceovers, deploy IndexTTS according to the README and fill in the service address

Risks

  • The README only covers installation on Windows 11; high-fidelity local voiceovers require a dedicated GPU and a separately deployed TTS model
  • It includes voice cloning features, which can only clone your own voice or voices for which you have obtained authorization; dependencies like Remotion installed with the package have their own license terms, which must be verified separately before commercial use

Open

GitHub

Explore the latest AI Radar updates