Automated AI Digital Human Video Generation Toolset Released on GitHub

A new open-source project titled digital-human-koubo by developer Berryxia (andyhuo520) has been published on GitHub, providing an agent-driven framework to generate talking-head presenter videos from a single photograph. The repository integrates remote MCP generation, automated quality assurance, voice spectral matching, and multi-style creator packaging.

Core Workflow and Capabilities

The project is structured as an Agent Skill (`SKILL.md`) designed to work with AI coding assistants and agent frameworks such as Claude Code, Codex, OpenClaw, and WorkBuddy. According to the documentation, the primary workflow operates through several automated stages:

• Persona Persistence: Stores a fixed presenter configuration (face, outfit, voice description, prompt template) in `persona.json` to maintain visual consistency across clips.
• Segment Generation: Splits scripts into segments under 15 seconds and utilizes video generation models via the Renoise Model Context Protocol (MCP) server.
• Quality Control: Automated checks use Whisper transcription to verify audio against the script, alongside frame inspection to detect and regenerate segments with rendering issues.
• Audio Unification: Applies spectral matching, loudness normalization, and reverb to ensure separately generated clips sound cohesive.

Creator Packaging and Styling

The repository includes five editing styles and 48 creator-grade packaging components reverse-engineered from popular creator formats. Key visual features include:

• Text-Behind-Presenter: Utilizes macOS Vision segmentation to place graphics between the background and the speaker.
• Automated Thumbnails: Generates a 16:9, 9:16, or 3:4 hook frame as the opening 0.5 seconds of the video.
• Multi-Language Support: Documented tests cover English, Japanese, Korean, French, Spanish, and Arabic.

Prerequisites and Execution

Running the pipeline requires a local environment with `ffmpeg`, Python, and Node, alongside an active account on Renoise for generation credits. An optional integration with Xiaomi MiMo TTS is available for custom or cloned voices.

Practical Implications and Verification

The author emphasizes ethical guidelines, requiring explicit permission for any face or voice used, mandatory disclosure labels identifying AI-generated content, and explicit user approval before credit expenditure. Readers can verify the original announcement, review setup instructions, and inspect the code directly on the GitHub repository.