shuyu-labs/BigBanana-AI-Director

BigBanana AI Director – One‑stop AI short‑film / motion‑comic platform

What it is – A web‑based workbench that lets creators turn a short story or a single sentence into a complete short‑film or motion comic. It stitches together large‑language‑model text generation, image generation, and video‑to‑video interpolation into an industrial‑style “Script → Asset → Keyframe → Video” pipeline.

Core ideas

• Keyframe‑driven generation – the system first creates a precise start‑frame and end‑frame for each shot, then uses the Veo interpolation model to fill the motion between them. This gives fine control over camera moves and guarantees that characters and props stay consistent.
• Project‑level workspace – all assets (characters, scenes, props, world‑building notes) are stored in a central hub and can be reused across episodes or even across different projects.
• Four production phases
  1. Narrative planning – AI parses a plot outline or novel excerpt into a structured script (characters, scenes, props, shot list). Users can edit or let the AI continue writing.
  2. Consistency assets – generates reference images for each character (“make‑up” shots), scene concept art, and reusable prop images. Assets are versioned and can be batch‑filled.
  3. Shot workbench – a grid UI shows every shot. For each shot you can upload or generate a start/end keyframe, preview nine candidate compositions, and run either image‑to‑video or start‑end interpolation.
  4. Delivery – a timeline editor (named CutOS) lets you reorder, trim, add filters and finally export the full video, a zip of clips, or raw assets for downstream editors like Premiere or DaVinci Resolve.
• Prompt management – a single page lists all prompt templates (for script, characters, scenes, keyframes, video) with history and rollback, making it easy to debug inconsistent outputs.

Technology stack

• Front‑end UI served from a Docker container (no local build required).
• Backend is a thin orchestrator that forwards calls to the AntSK API platform, which aggregates many large‑model providers:
  • Text: GPT‑5.2 / GPT‑5.1 / Claude 4.6 Sonnet
  • Image: Nano Banana Pro, Gpt Image 2
  • Video: Sora‑2, Veo‑3.1, Vidu, Seedance 2.0, happyhorse, etc.
• All data (project metadata, generated assets) are stored in the browser’s local storage for personal use; larger deployments can mount external volumes via the Docker compose file.

How to run (public version)

# clone the repo (contains only docs, docker‑compose and the official image reference)
git clone https://github.com/shuyu-labs/BigBanana-AI-Director.git
cd BigBanana-AI-Director

# start the official Docker image
docker‑compose up -d   # pulls the pre‑built image, creates a container exposing port 3005

# open a browser at http://localhost:3005 and follow the on‑screen guide to add your AntSK API key

Updates are delivered solely through the Docker image; the source code in the repo is kept static for reference only. Commercial customers can obtain the full source under a separate license.

Licensing – The public repository is released under CC BY‑NC‑SA 4.0 (free for personal, non‑commercial use, with attribution and share‑alike). Commercial use requires a paid license from the authors (contact antskpro@qq.com).

Who it’s for

• Independent creators who need an end‑to‑end pipeline for short‑form video or animated comics.
• Small studios that want to batch‑produce episodic content while keeping visual consistency.
• Anyone looking to experiment with AI‑generated storytelling without wiring together multiple APIs manually.

---

Built for creators, by BigBanana.