Anning01/novelvids

What it solves

Manually producing short-drama videos from novels is slow and labor-intensive: you have to split chapters, design characters and scenes, write per-shot prompts, generate images and video, and then stitch everything into a finished episode. This project automates almost the entire pipeline, letting you go from a text novel (or an existing video) to a production-ready short drama with consistent characters, voices, and storyboards.

How it works

The system takes a novel manuscript (paste text or upload doc/docx/txt/pdf) and runs it through a series of automatic steps: chapter splitting, entity extraction (characters, scenes, props), asset generation (reference images, voice clips), storyboard generation, multimodal video synthesis, and final episode merging. You can also upload existing videos (mp4/mov, up to 500 MB and 20 minutes) into the "remake workshop", where the system detects global roles and scenes, breaks the video into shots, and regenerates professional storyboard prompts.

Both an automatic Agent mode and a manual editing mode are provided. The Agent (built on Pydantic AI) can answer natural-language requests like "add a rainy-night station scene to this chapter" or "make this shot warm-toned but keep the actions and dialogue", and it returns auditable change records with undo support. The platform also includes a storyboard editor, an infinite canvas (Vue Flow) for visual workflow, a voice library for character voice consistency, configurable video model adapters (Seedance, MiniMax H3, Wan3), and a billing dashboard with usage and cost tracking.

Who it's for

Content creators, short-drama production teams, hobbyists, and developers who want an end-to-end AI-assisted pipeline for turning books or reference videos into serialized short dramas. It is designed as a production-grade, deployable application (Docker recommended) rather than a demo, and includes team features (optional), role-based access, and extensive customization hooks for AI prompts and generation rules.

Highlights

• One-click pipeline from novel to finished short drama: chapter split, entity extraction, asset generation, storyboard, video synthesis, episode merge.
• Agent-assisted creation via natural language with real business tools, long/short-term memory, and reversible operation receipts.
• Remake workshop: upload a video or a folder of episodes, automatically detect scenes and shots, and generate new storyboard prompts with async progress tracking.
• Consistent assets: character appearances, outfits, aging states, and voice references are managed per asset with version history.
• Storyboard engine with two built-in strategies (cinematic narrative and narration) and fully editable, asset-tagged prompts.
• Infinite canvas (Vue Flow) for storyboard workflow, with pan, zoom, copy/paste, undo/redo, and viewport persistence.
• Config-driven video factory supporting Seedance, MiniMax H3, and Wan3, with first/last frame continuity and FFmpeg-based fallback.
• Model-agnostic cost billing: token/image/video usage, pricing snapshots, discounts, and per-project dashboards.
• Fully customizable prompt templates (Chinese and English) for every generation stage, without touching code.
• Production-ready: Docker compose, SQLite/PostgreSQL, local media or Aliyun OSS, and team authentication as an optional switch.