cclank/lanshu-create-ai-presenter-video

What it solves

This project provides a generalized workflow for creating AI-powered digital human presenter videos. It automates the complex process of turning a topic or script and a reference image of a person into a fully edited video featuring a digital avatar that speaks the text with synchronized lip-syncing, subtitles, and visual effects.

How it works

The system operates as a "Skill" for Codex, coordinating several AI capabilities (voice generation, video generation, and lip-syncing) without being locked into a specific service provider. The process follows a strict timeline based on the audio track:

1. Preparation: It takes a script and a licensed reference image.
2. Generation: It generates the full voiceover, which serves as the master timeline for the entire video.
3. Production: It creates the digital human footage, ensuring lip-syncing is aligned with the audio.
4. Post-Production: It adds subtitles, keyword animations, and covers, then renders the final video using FFmpeg.
5. Quality Assurance: It performs technical and manual checks to ensure audio-visual alignment and quality before outputting the final master and sharing versions.

Who it’s for

Content creators and users of Codex-compatible agent environments who want to produce professional-looking digital human presentation videos without manually coordinating multiple AI tools.

Highlights

• Provider Neutral: Not bound to specific models or private APIs, allowing it to use whatever tools are available in the current environment.
• Audio-Centric Timeline: Uses the full voiceover as the primary time reference to prevent lip-sync drift and seamless transitions.
• Automated Pipeline: Handles everything from script organization and voiceover to editing, rendering, and QA reports.
• Safety and Cost Controls: Includes built-in checks for image/voice licensing and cost-estimation before paid generation.