AICON (ai-moive-studio) is an open-source, self-hostable AI video creation workbench that turns script-to-video production into an editable workflow of text, image and video nodes on an infinite canvas, driven by a natural-language assistant, for creators who want the whole pipeline from script breakdown to distribution instead of a single prompt box.

What it is

AICON is a Python project, licensed under Apache License 2.0, that organizes AI video creation as a graph rather than a chat turn. A natural-language assistant works on an infinite canvas, creating nodes and connections between text, image and video material, so a production stays editable and inspectable at every step. It ships as a self-hosted application: the README documents Docker and Docker Compose v2 deployment, with a hosted demo available at aicon-studio.com.

The concrete problem it solves is the single-input-box model of AI video generation, where a prompt produces a clip that cannot be reorganized, re-referenced or continued. AICON replaces that with a chain: script breakdown into characters, scenes and shots; character reference images, keyframes and transition videos; chapter-level images, voice and subtitles assembled into commentary videos; and finally video composition with distribution to Bilibili. The project lives in the Python and Docker self-hosting ecosystem, and sits alongside model providers rather than bundling them — model calls are billed by whichever provider the operator configures.

Key capabilities

• Script-to-storyboard breakdown that extracts characters, scenes and shots, then generates character reference images, keyframes and transition videos.
• Infinite canvas that connects text, image and video nodes into one editable graph.
• Natural-language canvas assistant that creates nodes and connections; generation tasks are triggered manually by the operator.
• Per-chapter illustrated content: images, synthesized voice and subtitles combined into commentary videos.
• Asset and history management for generated material and reference assets, with video composition and Bilibili distribution.
• Provider-agnostic model configuration through the built-in "API 密钥管理" panel, supporting custom OpenAI-compatible endpoints with your own API Key and Base URL.
• Docker Compose v2 deployment via docker-compose.prod.yml and an .env.production file.

Who uses it and how

• Creators and small studios turning existing prose or scripts into storyboarded video, using the canvas to keep characters and shots consistent across a long piece.
• Self-hosters who run their own infrastructure and want the creation graph on their own hardware rather than inside a vendor account.
• Producers distributing finished work to Bilibili, using the built-in composition and publishing path instead of exporting to separate tools.
• Teams without GPU capacity, who point the workbench at hosted model APIs — the README's sponsored route is MiniMax H3 through the 秘塔科技 API, OpenAI-protocol compatible and also usable from ComfyUI.
• Developers extending the system; the repository separates frontend/README.md and backend/README.md, with a user guide, model configuration guide and changelog in Chinese.

Getting started

Clone the repository, copy .env.production.example to .env.production and fill in the database, Redis, JWT and MinIO settings, then run docker compose --env-file .env.production -f docker-compose.prod.yml up -d and open http://localhost to register model credentials. A hosted trial is also offered at aicon-studio.com for anyone who wants to evaluate before deploying.

How it compares

The facts provided name no comparable products or paid alternatives, so on that evidence this project stands alone in this registry. ComfyUI appears only as an interoperability target for the sponsored video API, not as a competing workbench. Practically, AICON is Apache-2.0 and self-hosted, with no licence fee or per-project charge from the project itself; model inference is paid directly to whichever provider is configured, and the optional relay at api.aicon-studio.com is explicitly not required.

When to use it — and when not to

A self-hoster must operate a database, Redis, an object store (MinIO) and a JWT secret, and must supply model provider credentials whose usage is billed by that provider. It is not the right pick for someone who wants a fully managed, zero-infrastructure product, nor for anyone expecting the assistant to run generation unattended, since canvas assistance creates nodes and connections while generation tasks are triggered manually. Documentation and the README are primarily in Chinese, which raises the bar for non-Chinese-speaking operators.

git clone https://github.com/869413421/ai-moive-studio.git
cd ai-moive-studio

cp .env.production.example .env.production
# 编辑 .env.production，填写数据库、Redis、JWT、MinIO 等配置

docker compose --env-file .env.production -f docker-compose.prod.yml up -d