ArcReel Video Creation Workspace
A self-hosted agent workflow that turns fiction into storyboards and short-form video.
- Source repo
- ArcReel/ArcReel
- Stars
- ★ 6.2k
- Last updated
- today
- License
- AGPL-3.0
- Primary language
- Python
- FA score
- 52/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API · Claude API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- An independent creator adapting a long novel episode by episode, with approval before each planned section is physically split and produced.
- Not a fit if
- Users wanting zero-config use without Docker and multiple provider keys
- Organizations whose policies rule out AGPL-3.0 with extra NOTICE terms
- Source review
- 52/100 · Major gaps
What does this agent do, and when should you use it?
ArcReel is a self-hosted web workspace for turning a novel or completed screenplay into character and clue assets, scripts, storyboard images, video clips, and finished video. Its React 19 interface is backed by FastAPI, while its assistant runs on the Claude Agent SDK with an orchestration Skill and focused Subagents. ImageBackend, VideoBackend, and TextBackend provide a common layer over multiple generation providers, with paths for custom OpenAI-compatible and Google-compatible APIs. GenerationQueue schedules image, video, and audio work in separate channels with RPM limiting, lease-based scheduling, and resume support; the UI exposes live progress through SSE. Docker Compose supports a default SQLite deployment and a PostgreSQL production deployment, while FFmpeg can assemble output video and episode assets can be exported as Jianying drafts.
After a user uploads a novel, manga-workflow reads project.json and the filesystem to determine the project stage, then dispatches Subagents such as analyze-characters-clues, split-narration-segments, normalize-drama-script, and create-episode-script as needed. It produces character and clue reference images, storyboard or grid images, then sends work through configured ImageBackend, VideoBackend, and TextBackend providers; video generation can be image-to-video, grid-to-video, or reference-to-video. GenerationQueue processes media jobs, ProjectManager handles filesystem and version history, and UsageTracker records provider-specific usage. Generated clips can be assembled with FFmpeg or exported as Jianying 5.x/6+ draft ZIP files. External agent platforms such as OpenClaw can authenticate with a Bearer API key, call POST /api/v1/agent/chat, and retrieve the Skill definition from /skill.md.
- An independent creator adapting a long novel episode by episode, with approval before each planned section is physically split and produced.
- An animation team that needs character, scene, and prop reference images to maintain visual continuity across shots.
- A commerce-video team uploading product photos to create an eight-part sales-video script and a Jianying draft with narration subtitles.
- A producer with a finished screenplay who wants to preserve dialogue and voice-over while generating assets and video afterward.
- A self-hosting team that needs to switch providers globally or per project among Gemini, Volcengine Ark, Grok, OpenAI, Vidu, DashScope, MiniMax, Kling, or compatible custom endpoints.
- A team using an external agent platform such as OpenClaw to create projects, generate scripts, and produce video through natural-language requests.
How do you install or deploy this agent?
Deploy on Linux, macOS, WSL2, or Docker. Run:
git clone https://github.com/ArcReel/ArcReel.git
cd ArcReel/deploy
cp .env.example .env
docker compose up -dOpen http://localhost:1241. For PostgreSQL production deployment, enter deploy/production, copy .env.example to .env, set POSTGRES_PASSWORD, and run docker compose up -d. The initial username is admin; set its password with AUTH_PASSWORD in .env, or allow first startup to generate and write one back to .env.
How do you use this agent?
Sign in and open /app/settings. Under ArcReel Agent, configure credentials, Base URL, and model for an official Anthropic or compatible provider. Then configure at least one image, video, or text provider API key, or add a custom OpenAI-compatible or Google-compatible provider. Create a project, upload a novel, finished screenplay, or product images, choose narration, drama, or ad/short-film mode, and confirm the assistant’s stage summaries as the workflow advances. When storyboard and video jobs complete, preview assets, follow SSE task status, inspect version history, then assemble the final output with FFmpeg or export a Jianying draft.
What are this agent's strengths and limitations?
- Defines a concrete production path from novel or screenplay input through reference assets, storyboards, clips, FFmpeg assembly, and Jianying-draft export.
- Uses ImageBackend, VideoBackend, and TextBackend to normalize multiple providers while also accepting custom OpenAI-compatible and Google-compatible APIs.
- Separates orchestration from focused Subagents, keeping large source text inside the Subagent context and returning summaries for user confirmation between stages.
- Its queue has distinct image, video, and audio channels with RPM limiting, lease-based scheduling, and resume support.
- Supports global and project-level provider selection, version rollback, and provider-specific cost tracking and estimates.
- Generation requires API credentials for at least one external provider, and supported modalities, parameters, and pricing differ by provider.
- Production deployment requires PostgreSQL; even the default deployment depends on Docker, Docker Compose, and network access.
- On native Windows, Bash sandboxing and bwrap-based POSIX isolation are automatically downgraded; WSL2 or Docker Desktop is recommended for production.
- The project is licensed under AGPL-3.0 with additional terms in NOTICE, which may be incompatible with some organizational commercial-use policies.
- Only OpenClaw is explicitly documented as an external agent-platform integration; equivalent integrations for other agent platforms are not described.
How does this agent compare with similar options?
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| ArcReel Video Creation Workspace This agent | 52 · Major gaps | Self-hosted serviceFree + model costs | ★ 6.2k | today | Python | OpenAI API · Claude API |
| waoowaoo AI Film Studio | Insufficient evidence | Self-hosted serviceFree + model costs | ★ 15k | 1mo ago | TypeScript | — |
| OpenMontage Drama Studio | 47 · Major gaps | Self-hosted serviceFree + model costs | ★ 18 | 1d ago | Python | — |
| ReelMimic Style-Mimicking Video Studio | 50 · Major gaps | — | ★ 1.8k | 1d ago | JavaScript | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README mentions default admin password generation, AUTH_PASSWORD setting, warning about AUTH_ENABLED=false, and SECURITY.md notes about lack of multi-tenant isolation, indicating some consideration for permissions and sensitive data handling. However, no explicit least privilege design (e.g., minimal privilege roles), user confirmation mechanisms (e.g., confirmation before generation), data flow transparency (e.g., data flow diagrams), or details on encryption of sensitive data were found. For dependency security, CodeQL and Trivy scanning workflows exist, but no vulnerability remediation strategy or dependency locking details. External effects: third-party service dependencies are mentioned, but no explicit restrictions or audits. Rollback: README mentions version history and project archives, supporting rollback. Source attribution: LICENSE and NOTICE exist, but no explicit third-party code attribution. Deductions: lack of specific implementation details, mostly assertive descriptions.
Evidence shows: README and pyproject.toml specify dependency versions, test configuration (pytest markers) and CI workflow (test.yml) exist, indicating some self-consistency. However, no specific failure message examples or error handling strategies provided. Dependency availability: dependency list is complete, but no version compatibility or locking mechanism described. Deductions: static review cannot verify runtime behavior, and failure message details are missing.
Evidence shows: README clearly lists multiple workflows (novel adaptation, short videos, ads) and target users (creators, operators, contributors), capability boundaries (e.g., no multi-tenant support, partial Windows support) are described. Trigger precision: mentions Agent detects project stage and delegates to subagents, but no specific trigger conditions. Environment fit: supports Docker, SQLite/PostgreSQL, multi-platform images. Deductions: trigger precision description is vague, lacking specific rules.
Evidence shows: README provides detailed information architecture (documentation index, quick start, architecture diagram), clear installation instructions (Docker Compose, environment variables), stable naming (version managed by release-please), examples and FAQ exist, known limitations are explicit, license is AGPL-3.0, CHANGELOG and version management exist. Maintenance responsibility: CONTRIBUTING.md and community channels exist, but no explicit maintainer identity or responsibility statement. Deductions: maintenance responsibility description is unclear.
Evidence shows: README describes output usability (editable CapCut drafts, video export), marginal value (complete pipeline from content to final cut), cost-benefit (cost estimation and actual usage tracking). However, no actual user feedback or performance data. Deductions: static review cannot verify actual effectiveness, and quantitative metrics are lacking.
Evidence shows: README feature descriptions correspond to code structure (e.g., provider abstraction, queue), but no specific implementation links or test results provided. Cross-source corroboration: CI badges and Codecov links exist, but no specific data. Fact-inference separation: README distinguishes feature descriptions and limitations, but does not explicitly mark inferences. Deductions: lack of traceable specific evidence, mostly assertions.
- Static review cannot verify runtime behavior; all scores are inferred from file contents.
- No explicit user confirmation mechanism found; generation operations may execute automatically, configure with caution.
- Sensitive data handling details are insufficient; review code to confirm API key storage.
- Dependency security scanning exists, but no vulnerability remediation strategy; monitor dependency updates.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
[0.35.0](https://github.com/ArcReel/ArcReel/compare/v0.34.0...v0.35.0) (2026-11-01) 🌟 版本亮点 * 项目回收站： 在项目中删除的内容不再直接清除，会在项目回收站中保留 30 天，期间可以恢复；Agent 删除的内容同样如此。删除的整个项目会出现在项目列表页的「已删除项目」中，也保留 30 天。需要立即释放磁盘空间时，可以彻底删除。 * 文件视图： 新增文件视图，可以按目录查看项目中的全部文件和占用空间，并把不再使用的旧媒体移入项目回收站。 * Agent 审图： Agent 可以查看生成的资产图和分镜图，发现问题时建议重新生成。 * 稳定性： 大项目批量出片时，服务不再因卡顿崩溃重启。Docker 部署重启后，登录密码不再改变，已登录用户也不会掉线。 ✨ 新功能 …View on GitHub ↗ -
Release notes excerpt
[0.34.0](https://github.com/ArcReel/ArcReel/compare/v0.33.0...v0.34.0) (2026-10-24) 🌟 版本亮点 * 模型目录： 新增统一的模型目录。内置供应商、自定义供应商和 Agent 都从这里读取模型的能力和价格，并自动同步官方最新数据。所有模型集中在「供应商」页面管理，支持获取、添加、启用和停用，也可以自行修改模型信息和价格。如果原来的默认模型已经下架，升级后会自动换成当前在售的模型。 * 计费： 生成调用和 Agent 会话的参考费用改为按模型目录中的价格计算。查不到价格的调用会标为「未定价」，不再套用其他模型的价格。使用记录新增「生成调用」和「Agent 会话」两个视图。 * 升级提示： 自定义供应商模型之前手动填写的价格，升级后会被清除。如果实际价格与官方价格不同，可以通过倍率调整。 * 协议与端点： 新增 Anthropic Messages、OpenAI Responses、Gemini 视频等调用端点，以及百炼、Gemini 和 MiniMax 的语音合成。自定义供应商可以为每个模型单独修改请求地址，使用非标准路径的服务也能接入。 …View on GitHub ↗ -
Release notes excerpt
[0.33.0](https://github.com/ArcReel/ArcReel/compare/v0.32.0...v0.33.0) (2026-10-08) 🌟 版本亮点 * 界面整体改版： 全部页面基于 shadcn/ui 重做，布局、视觉和操作方式统一。窗口较小时，弹窗的标题和按钮也始终可见。 * 保存与离开提示： 编辑内容统一为先修改、再保存。离开页面或切换选中项时，如有未保存的修改会先提示，不再静默丢弃。 * 工作区与 Agent 面板： Agent 悬浮球改为顶栏开关，项目菜单改为可搜索的项目切换器。Agent 的工具调用等过程收为单行显示，提问时问题显示在输入框位置。 * 项目页面： 项目大厅改为海报卡片，最近有改动的项目排在最前；概览改为单列的故事设定页；分集视图改为分集目录与原文并排；剧集页头精简为两行。 * 分镜与资产： 分镜编辑与设定集重新排版，各类资产使用同一个详情面板编辑。项目资产新增删除入口，删除前会列出仍在引用该资产的集。 * 设置位置调整： 外部 Agent 接入改为设置页，原「API 密钥」改称「访问令牌」。GitHub 访问代理和官方服务开关移到市场的「设置」标签页。项目设置改为侧栏分页。 …View on GitHub ↗
FAQ
Do I need to bring my own model credentials?
Can I use an endpoint that is not one of the built-in providers?
/v1/models to discover available models and infers image, video, or text media types from their names.