An AI Film Studio That Runs on SQLite and API Keys

What it does

Huobao Drama is a full-stack TypeScript application that automates the entire short drama production pipeline. Feed it a sentence or script and it deploys AI agents to rewrite the story, extract characters and scenes, generate storyboards, assign voices, and render the final cut with subtitles via FFmpeg. It is essentially a web-based control room for mass-producing AI-generated episodes without opening an editor.

The interesting bit

The backend orchestrates five specialized Mastra agents—script_rewriter, extractor, storyboard_breaker, voice_assigner, and grid_prompt_generator—that pass work around like an actual film crew. It also acts as a universal adapter for both Chinese and Western AI vendors, routing requests to OpenAI, Gemini, MiniMax, Volcano Engine, Vidu, or Alibaba depending on whether you need text, images, video, or TTS.

Key highlights

• Agent pipeline: Five Mastra agents handle distinct production stages from script parsing to prompt engineering.
• Vendor agnostic: Supports OpenAI, Gemini, MiniMax, Volcano/Seedance, Vidu, and Alibaba across different media types.
• Full-stack TypeScript: Nuxt 3 frontend with pure CSS, Hono backend, Drizzle ORM, and better-sqlite3.
• End-to-end automation: Generates characters, storyboards, footage, voiceover, and final compositing in one flow.
• Docker-ready: Ships as a single merged image with FFmpeg bundled.

Caveats

• Non-commercial license: Released under CC BY-NC-SA 4.0, so building a business on it requires negotiating terms or using the separate commercial edition.
• API dependency: The fully automated pipeline depends on paid third-party APIs for LLMs, images, video, and TTS; your mileage will vary with budget.
• Orchestration, not models: The heavy generative lifting is done by external services and FFmpeg; this is a workflow and integration layer.

Verdict

Great for developers experimenting with automated video storytelling or running a small AI drama studio. If you need enterprise-grade asset management or want to avoid cloud API bills entirely, you will hit limits—though the Ollama integration lets you at least localise the LLM layer.