Moyin Creator: an Electron pipeline that turns a script into storyboards and Seedance 2.0 video prompts

The problem Moyin Creator targets: shot-by-shot drift in AI drama production

Generating one good image or one good clip is a solved demo. Producing forty shots in which the same character wears the same coat, stands in the same room and speaks with a consistent lip shape is not, and that gap is what this project addresses. The README describes five chained panels (script, character, scene, director, and an S-tier panel for Seedance 2.0 and SkyReels-V4), with the stated property that each step's output flows into the next without manual stitching.

The intended user is not a solo illustrator. The repository description says the tool is built for batch production of short drama, anime series and trailers, and the feature list backs that up with a parallel task queue and automatic retry of failed jobs. If you make one clip a week, the pipeline overhead costs more than it saves. If you make thirty, the character-consistency layer is the reason to look at this at all.

One caveat about scope: the README describes the tool as production-grade, but the package.json version is 0.2.3 while the release list shows V0.2.9, and the download section still points at a packaged build labelled 0.1.7. Treat version numbers in the repository as inconsistent and check which one you are actually running.

How the pipeline actually moves data between panels

Architecturally this is an Electron 30 desktop app with a React 18 and TypeScript renderer, built through electron-vite on Vite 5, with Zustand 5 holding global state and Radix UI plus Tailwind CSS 4 for the interface. The AI logic is not in the renderer: the README's architecture table names an internal package, @opencut/ai-core, responsible for prompt compilation, the character bible and task polling.

The repository layout confirms the split. electron/main.ts handles storage, the filesystem and protocol handling, electron/preload.ts is described as the security bridge, and src/packages/ai-core sits alongside src/stores and src/lib. That means the renderer talks to the main process over the preload bridge rather than calling providers directly, and prompt compilation happens inside the app before any request leaves the machine.

The consistency mechanism is the piece worth understanding. The README calls it a six-layer identity anchor and pairs it with a Character Bible plus bound reference images. In practice, the S-tier panel collects character reference images, scene images and first-frame images automatically, then fuses three layers into the prompt: action, camera language, and dialogue with lip sync. First frames are assembled by grid stitching using what the README calls an N×N strategy. None of this is a model-side feature; it is prompt and asset assembly done locally.

Installing Moyin Creator and running a first script through it

The README lists Node.js 18 or newer and npm 9 or newer as the only environment requirements. The install path is the standard clone, install, dev loop, and the README gives exactly these commands.

# 克隆仓库
git clone https://github.com/MemeCalculate/moyin-creator.git
cd moyin-creator

# 安装依赖
npm install

# 启动开发模式
npm run dev

After the dev server starts you should see the Electron window with the five panels. The next step is not a config file. The README says to open 设置 → API 配置 (Settings, then API configuration) and enter your AI provider's API key there. There is no documented .env file, no documented config key, and no documented way to provision keys headlessly, so a first run is a GUI task.

Building a distributable is where the scripts diverge. The README shows a generic build, but package.json defines platform-specific targets through a Node wrapper, and the prebuild step is a PowerShell script.

# 编译 + 打包 Windows 安装程序
npm run build

# 仅编译（不打包）
npx electron-vite build

For other platforms package.json exposes build:win, build:mac, build:linux, and macOS variants including build:mac:universal. Note that prebuild runs scripts/prebuild-cleanup.ps1 through powershell, which is a Windows-first assumption in a cross-platform build script. The README also links a packaging download on Baidu Pan with extraction code 8888, labelled as version 0.1.7 corresponding to the open-source code.

Seedance 2.0 constraints are validated locally, and that is the sharpest feature

The S-tier panel does something concrete that is easy to overlook: it checks Seedance 2.0 parameter limits before submission. The README states the constraints as no more than 9 images, no more than 3 videos, no more than 3 audio files, and a prompt of at most 5000 characters.

That matters because multi-shot narrative generation means merging several storyboards into one request, and merging is exactly how you exceed a modality budget. A tool that assembles the request and then validates it against the provider's ceiling saves a failed round trip per offending shot. The README also lists multi-modal references through @Image, @Video and @Audio syntax, with character reference images, scene images and first frames collected automatically rather than attached by hand.

There is a real limit here that the documentation does not resolve. The constraint set is specific to Seedance 2.0, and the panel also mentions SkyReels-V4. The README does not explain how validation behaves for any other provider, nor whether the 9/3/3/5000 numbers are configurable. If you route through a different video model, assume you are on your own for budget checking until you read the source.

Where Moyin Creator is the wrong choice

The licence is the first hard boundary. The project is AGPL-3.0, and the README describes a dual-licence model: open-source use under AGPL-3.0, where modified code must be released under the same licence, and a separate COMMERCIAL_LICENSE.md for closed-source use or integration into a commercial product. If your business model depends on shipping a modified Moyin Creator without publishing your changes, the open-source path does not fit and you are in a negotiation, not a download.

The second boundary is that this is a local desktop application, not a service. The README documents no Docker image, no server deployment and no headless mode. A studio that wants a shared render farm with a web front end is not the target user, and nothing in the repository layout suggests that is planned.

The third is dependency on external providers. The README lists multi-provider scheduling and API key rotation with load balancing, but the app itself does not generate images or video. Every capability in the feature list is a wrapper around someone else's paid endpoint. If your provider's API is not among those the scheduler supports, the batch queue has nothing to call.

Finally, the update mechanism deserves scrutiny. package.json sets updateConfig.manifestUrl to an HTTP address on a bare IP, not HTTPS, alongside defaultGithubUrl and a Baidu Pan fallback. The README does not document rollback, nor what the manifest contains. That is a supply-chain surface worth reviewing before you deploy this on machines you care about.

How it differs from a general-purpose AI video editor

A tool like OpenCut, or any general timeline-based AI video editor, treats the model as a clip source: you generate footage, then you arrange it on a track and cut. The unit of work is the clip, and consistency between clips is the editor's problem to solve by hand through reference frames and retries.

Moyin Creator inverts that. The unit of work is the script. Parsing produces scenes, storyboards and dialogue; character identity is anchored before any image exists; the storyboard panel carries cinematography parameters (shot size, camera position, movement) as structured fields rather than as free text; and the final stage compiles everything into one Seedance 2.0 request covering multiple shots. The README's own framing is that the five panels are interlocking and each output feeds the next.

The practical difference shows up in what you can automate. With a timeline editor, batch generation means scripting the editor's API or driving it manually. With this pipeline, the README describes a one-click full flow from script parsing through character and scene generation, storyboard cutting, batch image generation and batch video generation, with a parallel queue and automatic retry. The trade-off is flexibility: a timeline editor lets you fix one bad shot in isolation, while a script-driven pipeline pushes you toward regenerating a group.

Maintenance, version drift and what upgrading costs

The repository is not archived and the last push was on 2026-08-27, so the project is being worked on. The release cadence shown in the release list is fast and uneven: V0.2.9 on 2026-06-17, v0.2.8 on 2026-05-25, v0.2.7 on 2026-05-10, each labelled a major update. Three releases in about five weeks is a pace that implies API surfaces and prompt assembly rules can move between versions.

That pace has a cost for anyone building on top. The open-source package.json declares version 0.2.3 while releases reach V0.2.9, and the README's download section still points at a 0.1.7 package. A CHANGELOG.md and a CHANGELOG-v0.1.8.md both exist at the repository root, so the history is documented, but the version strings are not reconciled in what the README shows. Before upgrading a production install, read CHANGELOG.md rather than trusting the version in the interface.

On licence cost: AGPL-3.0 means network use of a modified version triggers source disclosure obligations, and the README routes commercial closed-source use to COMMERCIAL_LICENSE.md. This is a summary of what the README states, not legal advice; the terms that bind you are in LICENSE and COMMERCIAL_LICENSE.md.

Editorial conclusion

Adopt Moyin Creator if you already have API keys for image and video providers and you produce episodic short drama or anime in volume, because the value is in the batch queue and the Seedance 2.0 constraint checking, not in the editor. Do not adopt it if you need closed-source redistribution (the licence is AGPL-3.0 with a separate COMMERCIAL_LICENSE.md), or if you want a browser tool with no local install. Verify first that your provider's API is one the app can schedule, that a full build on your platform works through scripts/build-desktop.mjs, and that the update manifest host in package.json is one you are willing to contact.

Frequently asked questions

Official sources

1. License: AGPL-3.0
2. MemeCalculate/moyin-creator on GitHub
3. Project website
4. README
5. Releases