Museboard

Museboard is a macOS-first desktop app for visual creation planning. It connects a structured mind map, an image reference canvas, local LLM text generation, and local ComfyUI image generation into one workflow.

Museboard 是一个面向绘画、插画、角色设计和场景设计的桌面工具。它把“创作文案、思维导图、参考图管理、AI 深化节点、本地 ComfyUI 生图”放在同一个工作区里。

Current Status

Museboard is an MVP. It is usable for local testing, but public distribution is not fully finished yet.

• Platform priority: macOS
• Desktop framework: Tauri 2
• Frontend: React + TypeScript + Vite
• Text model integration: LM Studio, OpenAI API, DeepSeek API, Ollama, and custom OpenAI-compatible endpoints
• Image generation integration: local ComfyUI, with Flux-oriented preset support
• Distribution status: local unsigned macOS app/DMG only

Public macOS release still requires Developer ID signing and notarization.

Main Features

• Node-driven mind map for visual ideation
• Creation templates for story illustration, character design, object design, and scene design
• AI-assisted node expansion: select any node and generate the next level of child nodes
• AI-generated node descriptions and final briefs
• Canvas reference management linked to mind-map nodes
• Parent nodes aggregate images from all descendant nodes
• Per-node canvas layout, so the same image can have different position, scale, rotation, mirror, and black-and-white display states in different views
• Drag-and-drop local images or image links
• Multi-select box for moving images and annotations together
• Annotation tools: rectangle, arrow, text, and freehand line
• Clean canvas mode for reference-only viewing
• Always-on-top mode for drawing alongside other apps
• Local ComfyUI image generation, including Flux preset setup and generated-image import back into the root node
• Chinese and English UI/text-generation language switching

Typical Workflow

1. Create or open a Museboard project.
2. Choose a creation type, such as story illustration or character design.
3. Build the mind map manually, or use AI to generate and deepen nodes.
4. Generate a brief from confirmed nodes.
5. Click any node and drag reference images into its canvas.
6. Use parent nodes to review all images from child nodes together.
7. Add simple annotations, mirror images, or switch images to black and white when comparing values and silhouettes.
8. Optionally send the brief to ComfyUI, generate an image, and save the result back into the top-level node.

Canvas Controls

• Scroll wheel / trackpad scroll: pan around the canvas.
• Cmd + scroll or trackpad pinch: zoom around the pointer.
• Cmd + =: zoom in.
• Cmd + -: zoom out.
• Cmd + 0: reset the canvas view.
• Tab: toggle clean canvas mode, hiding or restoring extra UI around the canvas.
• Hold Space and drag: pan the canvas, matching ArtFlex-style navigation.
• Tap Space: fit the selected image group; if nothing is selected, fit all visible images. Tap Space again to return to the previous view.
• Double-click an image: toggle between its normal size and a 75 x 75 thumbnail.

User Setup

Install From Source

Requirements:

• macOS
• Node.js 20+
• Rust toolchain
• Xcode command line tools

Install dependencies:

npm install --cache .npm-cache

Run the browser preview:

npm run dev

Open:

http://127.0.0.1:5173/

Run the desktop app in development:

npm run tauri -- dev

Build the macOS app and DMG:

npm run tauri -- build

Generated local release artifacts:

• src-tauri/target/release/bundle/macos/Museboard.app
• src-tauri/target/release/bundle/dmg/Museboard_0.1.0_aarch64.dmg

Text Model Setup

Use Settings -> Text Model.

Supported providers:

• LM Studio: local OpenAI-compatible endpoint, default http://localhost:1234/v1
• OpenAI API: default https://api.openai.com/v1, requires an API key
• DeepSeek API: default https://api.deepseek.com, requires an API key
• Ollama: native local endpoint, default http://localhost:11434
• OpenAI-compatible: custom endpoint for other compatible servers

API keys are saved only in local app preferences. They are not written into project.museboard.json.

LM Studio

http://localhost:1234/v1

In LM Studio:

1. Load a chat model.
2. Start the local server.
3. Copy the exact model name.
4. Open Museboard settings.
5. Paste the endpoint and model name into the Text Model section.

OpenAI API

1. Create an API key in your OpenAI account.
2. In Museboard settings, choose OpenAI API.
3. Keep endpoint as https://api.openai.com/v1.
4. Paste the API key.
5. Enter the model name you want to use.

DeepSeek API

1. Create a DeepSeek API key.
2. In Museboard settings, choose DeepSeek API.
3. Keep endpoint as https://api.deepseek.com.
4. Paste the API key.
5. Use a current DeepSeek chat model name, such as deepseek-v4-flash, unless DeepSeek changes its model list.

Ollama

1. Install Ollama and pull a chat model.
2. Start Ollama.
3. In Museboard settings, choose Ollama.
4. Keep endpoint as http://localhost:11434.
5. Enter the local model name, for example qwen3:8b.

The selected text model powers:

• random mind-map generation
• missing-node completion
• node-description generation
• child-node generation
• brief generation
• prompt rewriting for image generation

ComfyUI Setup

Use Settings -> ComfyUI Setup Wizard.

Recommended flow:

1. Keep the endpoint as http://127.0.0.1:8188.
2. Click Test Connection.
3. If ComfyUI is not running, set the ComfyUI folder and launch command.
4. Click Start and Connect.
5. Click Flux Recommended Preset.
6. Click Generate Test Image.
7. If the test image appears, the main Generate Image button can use ComfyUI.

Detailed guide: docs/comfyui-setup.md

Project Files

Museboard saves each project as a folder:

project.museboard.json
assets/

project.museboard.json stores:

• mind-map nodes
• generated or edited brief text
• canvas layouts
• annotations
• image links
• project metadata needed to reopen the board

Runtime text-model and ComfyUI settings are stored in local app preferences, not in project files. Imported images are copied into assets/. Desktop projects load canvas images through generated previews so large boards do not keep full base64 images in memory.

Development

Run checks:

npm run check

Build a local release after checks:

npm run release:local

Known local toolchain limitation: cargo fmt and cargo clippy require Rust components that may not be installed. Install them with:

rustup component add rustfmt clippy

Documentation

• GitHub publishing guide
• Text model setup guide
• ComfyUI setup guide
• Manual test plan
• Release checklist

Known Limitations

• macOS public distribution is not signed or notarized yet.
• Windows and Linux packaging have not been verified.
• ComfyUI compatibility depends on the user's installed models and custom nodes.
• Browser preview cannot auto-start local ComfyUI.
• The app currently stores project data in a local folder format, not in a cloud sync format.

License

Museboard is open source under the MIT License.