ArcReel is a self-hosted, AGPL-3.0 AI video production workspace that turns novels, finished scripts, or product material into character-consistent short videos — characters, scenes, props, storyboards, clips and Jianying drafts — aimed at creators, small studios and technical teams willing to run their own Docker deployment.

What it is

ArcReel is an open-source, self-hostable workbench for AI video production, written in Python and published under AGPL-3.0 with additional terms in NOTICE. It organizes content analysis, asset management, storyboarding, media generation, cost tracking and export into one auditable production pipeline that can be paused and resumed. The repository lists 4,976 stars, 982 forks and 97 open issues, with the last push on 2026-09-18, and the project maintains a documentation site at docs.arc-reel.com alongside a homepage at arc-reel.com.

The concrete problem it solves is the gap between raw source material and an editable draft. It replaces the manual, tool-hopping step of reading a novel or script, hand-drawing or re-prompting characters, assembling shot lists, generating clips and then rebuilding everything inside a video editor. ArcReel takes novel, finished script or product material through content analysis and project planning, character/scene/prop assets, episodes and a structured script, storyboard images or multi-grid storyboards, video clips and voiceover audio, then out to final composition or a Jianying draft. Because asset images and other reference images are reused across storyboards, characters stay visually consistent from shot to shot, and because the export lands as a Jianying draft, the result remains editable rather than locked into a render.

Key capabilities

• Unified production chain from novel, finished script or product material through to characters, scenes, props, storyboards, video clips and a finished film.
• Cross-shot consistency by reusing asset images and similar reference images across storyboards.
• Human review at key stages: any single asset can be regenerated and historical versions can be rolled back.
• Agent-orchestrated or user-driven execution, with each stage available for review, adjustment or regeneration in the workbench.
• Unified configuration of Agent, text, image, video and TTS providers from the Settings page.
• Cost visibility, showing projected cost and actual usage before and after generation.
• Jianying draft export targeting the mainland China version of Jianying, for continued work on subtitles, dubbing, pacing and transitions; CapCut compatibility has not been verified.

Who uses it and how

• Creators working on AI comic-drama and novel adaptations, where one source text must yield consistent characters across many shots.
• Producers of narration or commentary short videos, and of advertising or product-selling clips built from product material.
• Operators running the default Docker Compose stack on SQLite, who can later follow the documented migration path to PostgreSQL for production loads.
• Teams with a review stage, where an operator approves key stages, regenerates individual assets and rolls back versions rather than accepting a single blind render.
• Developers extending provider adapters, tests and documentation, installing the project pre-commit hooks with uv run pre-commit install.

Getting started

Docker and Docker Compose are the only prerequisites: clone https://github.com/ArcReel/ArcReel.git, change into ArcReel/deploy, copy .env.example to .env, and run docker compose up -d. The workbench is then reachable on port 1241 with the default username admin; when AUTH_PASSWORD is left empty, the first startup generates a password and writes it back into deploy/.env.

How it compares

This registry's facts name no paid products that ArcReel replaces, and no direct peer project either, so it stands alone here. Jianying and CapCut appear only as export destinations, not as alternatives. The nearest practical contrast the facts support is with the closed editing workflow it feeds into, which is commercial software and not self-hosted.

When to use it — and when not to

A self-hoster must operate the application stack plus its data layer — SQLite by default, PostgreSQL optionally, with documented upgrade, backup and reverse-proxy procedures — and must arrange authentication and TLS before any remote access. Note that the default Compose file publishes port 1241 on all host network interfaces; do not expose the service directly to the public internet. Configure authentication first, and reach it remotely only over HTTPS, a VPN or a secure tunnel. Organizations that cannot accept AGPL-3.0, or that want commercial deployment, white-labeling or redistribution without the AGPL obligations, should not adopt it as-is and should contact [email protected] instead. It is also a weaker fit for teams that need CapCut output, since only mainland-China Jianying export is verified, and for anyone who will not accept a Chinese-language README.

flowchart LR
    A["小说 / 成品剧本 / 商品素材"] --> B["内容分析与项目规划"]
    B --> C["角色 / 场景 / 道具资产"]
    C --> D["分集与结构化剧本"]
    D --> E["分镜图 / 多宫格分镜"]
    E --> F["视频片段 / 旁白音轨"]
    F --> G["成片合成"]
    F --> H["剪映草稿导出"]

git clone https://github.com/ArcReel/ArcReel.git
cd ArcReel/deploy

cp .env.example .env
docker compose up -d

uv run pre-commit install