ArcReel
ArcReel/ArcReel
ArcReel is an open-source app project by ArcReel on GitHub: Self-hostable video workspace: novel/script → character, scene and prop assets, storyboard, video and Jianying (CapCut) drafts, with cross-shot consistency and cost tracking. It covers adaptation, characters, storyboard and production. 5.5k stars, last updated 2026-10-16, AGPL-3.0 license.
Editor's notes
ArcReel deploys with Docker Compose and is served on port 1241; after logging in as admin you set up text, image, video and TTS providers on the Settings page. An external agent first runs setup-arcreel-skills to connect to ArcReel's MCP service, then video-workflow advances the project by the next step that get_workflow_plan returns. Jianying drafts target the mainland China edition, and the README says CapCut compatibility is untested. The skills under .agents/skills are for developing the repo itself and have nothing to do with video production.
- Type
- Related open-source platforms (full apps, not skills)
- Stages
- Adaptation · Characters · Storyboard · Production
- Format
- Full application
- Language
- Multilingual
- GitHub stars
- ★ 5.5k
- Last updated
- 2026-10-16
- License
- AGPL-3.0
- SKILL.md files
- 11
- Checked by hand
- 2026-10-13
In the author's words
Open-source, self-hostable video workspace driven by AI agents for stories, storyboards and short-form video: turns novels and scripts into character, scene and prop assets, storyboards, videos and Jianying drafts, with cross-shot consistency, multiple providers and cost tracking.
Translated from the Chinese original.
Skills in this repo (11)
Names and descriptions come from each SKILL.md frontmatter; Chinese descriptions are shown in our English translation.
- adapt-custom-endpoint
Use when the user asks to connect a video or image provider that uses a JSON submit-then-poll protocol to ArcReel, or to write, validate, test or save a custom endpoint definition.
- edit-video
Edits one episode on the editing timeline and renders the final cut or exports a Jianying draft on request. Use when all of an episode's videos are ready and a cut is needed; also when the user asks for changes by clip number (e.g. "c3 drags"), or asks to add BGM, render the final cut, export a Jianying draft, copy or rename an editing timeline, or go back to an earlier revision.
- generate-assets
Unified asset generation skill: accepts `--type=character|scene|prop|product`, or with no type scans every pending resource (missing its sheet) and dispatches by type. Use when the user says "generate character image" / "generate scene image" / "generate prop image" / "generate product image", wants reference images for a new asset, or an asset is missing its *_sheet.
- generate-grid
Generates grid storyboard images. Use when the user says "generate grid", "grid image generation" or "assemble storyboard frames with a grid". Groups shots by segment_break, picks the best grid size and generates a combined grid with chained transition frames; after the user reviews and approves the combined image, it is cut into the starting frame for each shot.
- generate-narration-audio
Generates narration voice-over (TTS) shot by shot for narration/commentary scripts. Use when the first automatic cut of a TTS project needs missing voice-over filled in, when the user asks to generate or regenerate narration for a shot or episode, or when a batch voice-over run was interrupted and needs completing.
- generate-script
Prompt writing: calls the project's configured text model to fill in image_prompt and video_prompt for shots / video units still to be written in the final script (rewriting the unit body for reference-to-video); for ad projects without a final script yet, it generates the whole thing. Called by the create-episode-script subagent; its input is the final script itself and project.json.
- generate-storyboard
Generates storyboard images for shots. Use when the user says "generate storyboard" or "preview storyboard frames", wants to regenerate some storyboard images, or some shots in the script lack a storyboard image. Keeps characters and visuals continuous automatically.
- generate-video
Generates video for shots or self-contained video units. Use when the user asks to generate or redo video; supports a whole episode, single items and batch selection.
- manage-project
Project management toolkit. Used for adding or changing characters/scenes/props in project.json (via the patch_project tool, upsert by table+name), cascading asset renames (rename_asset), merging assets registered twice for the same identity (merge_asset), writing top-level settings fields, editing the project overview, and querying video model capabilities (get_video_capabilities). Episode planning is not in this skill: it goes through the mcp__arcreel__plan_episodes / reset_episode_planning server tools.
- setup-arcreel-skills
Connects the current agent host to ArcReel's remote MCP service and verifies access.
- video-workflow
Orchestrates ArcReel video projects when the user asks to create a video, start or continue a project, move to the next step, check progress, finish or export. For end-to-end workflow requests, not for editing a single existing artifact.
Similar projects
Alibaba's LumenX Studio: all-in-one production of short manju from novel to motion video, with script analysis, character customization, storyboard and video compositing
Automated generation tool for AI video, anime, short drama and manju
Toonflow: open-source AI short drama / manju platform with script, smart storyboarding, character and scene assets, image and video generation, on an infinite canvas with agents
Huobao Drama: generates a complete short drama from one sentence, fully automated from script to final video
AI film and video agent workspace (preview): image and video creation, from shorts to controllable live-action output