Three-Minute Book Reading Video Generator

The Problem

Turning a book into a short video often gets stuck on three parts: copy that sounds natural, stable image generation, and audio-subtitle alignment. book-video-generator automates that pipeline: it takes book_name and author_name, then outputs a roughly three-minute MP4 explainer video, reducing manual scripting, image retries, and per-line caption work.

Core Workflow

The skill splits the job into five stages:

• Copy generation: it uses the current platform's web search to collect the book's intro, publication year, and common readings, then asks the LLM for a 700-1000 word script.
• Scenes and chapters: the script is split into 8-50 scenes, each with subtitle text, visual description, and image prompt; four chapter titles are also generated for the top progress bar.
• Asset generation: each scene gets a 1024x768 flat illustration via ImageGen, Volcengine Jimeng, Gemini, or Agnes; subtitles are spoken with Volcengine TTS, falling back to edge-tts without credentials.
• Composition: compose_video.py uses ffmpeg to combine images, audio, ASS subtitles, cover frame, BGM, and transition sounds into a 1920x1080 video, with keyword highlighting and subtitle entrance/exit animation.

Boundaries

• Image and TTS generation depend on external models; CLI environments need the matching API key, MCP, or script setup.
• Volcengine TTS word timestamps are estimated from audio duration, while edge-tts native timestamps are usually more precise.
• BGM and page-turn audio may not be bundled; the video still renders without them if the files are missing.
• It is aimed at fixed-style book explainer shorts, not real-person narration, complex camera work, or multilingual dubbing.