book-video

Book Video is an open source workflow for making short, atmospheric book promotion videos using natural language instead of code or commands. It is built to run inside Codex, an AI coding assistant, and turns picking a book, writing a script, generating mood images, timing a voiceover, adding subtitles, and mixing background music into one repeatable process.

To use it, you open Codex and simply describe the book video you want in plain language. The first time you run it, Codex checks and sets up your local environment on its own, asking you directly whenever it needs a decision from you. From there, the steps look like this: you tell Codex which book to cover, or ask it to suggest five books that would work well as short videos. Codex confirms the title, author, and edition, then writes out a voiceover script as a code block you can copy straight into CapCut, while also saving it as a local script file. You review and approve the script, after which Codex generates the mood images and video frames. You record or export the narration as an MP3 using CapCut or another tool and give Codex the file path. Codex then uses speech recognition to build timing references, aligns subtitles against the script file as the source of truth, mixes in background music, and renders the final MP4. If you want the script, images, or audio changed, Codex generates a new version and swaps it in once you approve it.

The repository ships with four default background music tracks for learning and reference use. The project's code, docs, and templates are released under the Apache 2.0 license, but that license does not cover the bundled music, fonts, third party tools, or any AI generated images, so anyone reusing those assets needs to confirm their own rights separately. More detailed internal production rules live in the repository's AGENTS.md and playbook documentation files.