binwu1 /drama-series-agent

About drama-series-agent

Drama Series Agent takes a one-line idea and walks it through series settings, an episode script, cast portraits and the video render. Rendering runs on your own ComfyUI with MiniMax H3, and each shot starts from the last frame of the one before. If a render stops, you resume from that shot; if you dislike one shot, only that shot re-renders. You install ComfyUI, the H3 nodes and every model file yourself.

Best for: Creators with a local GPU and ComfyUI who want to turn a series idea into episode after episode of MiniMax H3 video.

• Writes series settings (plan, world, characters, art style, episode map), then each episode script
• Builds the shot table for ComfyUI and binds which cast member appears in which shot
• Chains shots: the last frame of one becomes the first frame of the next, no manual editing
• Resumes a stopped episode from the breakpoint; re-renders a single shot and re-joins the episode
• Only re-runs the steps a change actually affects, not the whole episode
• Web workbench (chat plus render jobs) or a headless drama-series-render CLI

1. 1
   Install ComfyUI yourself, add the MiniMax H3 Ref2VA custom nodes, and download every model file the workflows in workflows/selfhost/ name.
2. 2
   Clone the repo: the command below then cd drama-series-agent.

   git clone https://github.com/binwu1/drama-series-agent.git

   Copy
3. 3
   Make a Python 3.10+ venv and run the command below, then copy .env.example to .env and set COMFYUI_URL and your LLM keys.

   pip install -e ".[render,dev]"

   Copy
4. 4
   Check the install with python scripts/doctor.py. You want [OK] for ffmpeg, skills and the API.
5. 5
   Start it with ./start.sh --with-web (Windows: .\start.ps1 -WithWeb) and open http://localhost:5173/.