Build powerful ComfyUI custom node packs with a visual GUI and AI.
🌐 Landing page / 项目主页: https://caoool.github.io/comfyui-node-canvas/
⭐ Star this repo if it saves you ComfyUI boilerplate: https://github.com/caoool/comfyui-node-canvas
At a glance: local GUI app, visual node contract editor, editable generated Python, AI-assisted node creation, validation, ZIP export, and managed deploy into ComfyUI.
一句话： 这是一个本地 GUI 应用，用 AI 帮你更快创建、修改、验证、部署和测试 ComfyUI 自定义节点包。
Demo: a user asks for a simple LLM prompt-enhancer node; AI creates the node contract and code, updates dependencies, deploys the pack, and the node is tested in ComfyUI.
演示： 用户提出一个简单的 LLM 提示词增强节点需求；AI 创建节点接口和代码、更新依赖、部署节点包，并在 ComfyUI 中测试节点。
- Quick Start: clone, install, run, and create your first node.
- Wiki / detailed guide: setup, workbench, node creation, deployment, project structure, and troubleshooting.
- GitHub Pages landing page: short bilingual product page with screenshots.
- Node ideas: suggest a ComfyUI node or pack for the AI Builder.
- Feature requests: request GUI, AI, deploy, or validation improvements.
- Bug reports: report broken behavior with reproduction steps.
Use these prompts in the AI Builder to see what the project is designed for:
- "Build a node that takes a simple prompt string and turns it into a stronger image-generation prompt with a small LLM."
- "Create a utility node that reads image metadata and returns seed, model name, and prompt text."
- "Add a batch helper node that takes a folder path and outputs sorted image file paths."
- "Explain why this generated ComfyUI node fails validation and patch the Python file."
- "Add requirements and an install script for this node pack, then validate before deploy."
Stars help ComfyUI node authors find the project, help prioritize AI-assisted node-building features, and make it easier to collect real node ideas from the community.
ComfyUI Node Builder is a local GUI app for building ComfyUI custom nodes and full node packs with ease. It gives you a visual node contract editor, generated ComfyUI pack files, direct Python editing, AI-assisted authoring, local checks, ZIP export, and managed deployment into your ComfyUI installation.
Use it when you want to move from an idea to a working ComfyUI custom node without rebuilding the same pack structure by hand every time.
It is designed for:
- ComfyUI node authors who want a faster edit-test-deploy loop
- workflow builders prototyping image, latent, mask, audio, text, or utility nodes
- Python developers who want generated boilerplate without losing source control
- AI-assisted prototypers who want AI help, but still need visible file changes and validation
- anyone building more complicated ComfyUI packs with dependencies, install scripts, Return UI, or runtime frontend displays
ComfyUI custom nodes are powerful, but the authoring loop is repetitive:
- define the Python class and
INPUT_TYPES - keep return types, return names, mappings, and categories in sync
- add widgets, custom ports, and Return UI displays
- manage
requirements.txt,install.py, and shared files - restart ComfyUI after each deploy
- debug generated files when something drifts
This project turns that loop into a structured local workbench.
The AI Builder is built into the workbench, not bolted on as a separate chat page.
It can help you:
- create packs and nodes from a prompt
- revise selected node code and generated files
- explain validation or terminal errors
- validate the generated pack
- run terminal checks in the pack workspace
- request deployment when Allow deploy is enabled
- use your local Codex plan through the installed
codexCLI, or provider-backed models from OpenAI, OpenRouter, OpenAI-compatible servers, Anthropic, Gemini, or Ollama
Deployment is gated. The assistant can ask for a deploy action, but the UI only emits deploy when you explicitly allow it.
- Visual node contract: add inputs, outputs, widgets, custom types, Return UI, categories, and metadata.
- Full Python editing: edit generated node Python directly in a Monaco-powered workspace.
- Generated pack files: inspect node files,
__init__.py, runtime UI scripts, custom renderers, requirements, install scripts, and shared files. - Managed deployment: export ZIPs or write builder-managed packs into
<ComfyUI>/custom_nodes/ComfyUINodeBuilder/<pack-folder>/. - Round-trip metadata: keep
builder.project.jsonso builder-owned packs can be loaded back into the app. - Local helper server: file system writes, AI provider calls, terminal commands, dependency install, and ComfyUI restart requests stay on loopback.
Requirements:
- Node.js 22 or newer is recommended.
- A local ComfyUI checkout is required for direct deploy/load behavior.
- ComfyUI Extension Manager is recommended if you want restart requests after deploy.
Install and run:
git clone https://github.com/caoool/comfyui-node-canvas
cd comfyui-node-canvas
npm install
npm run devThe dev script starts:
- Vite frontend, usually at
http://localhost:5173 - helper server at
http://127.0.0.1:3001, proxied through/helper
Open the Vite URL, then open Settings from the status bar and configure:
- ComfyUI URL: usually
http://127.0.0.1:8188 - ComfyUI Install Path: the folder that contains ComfyUI's
custom_nodesdirectory - Pack Name / Pack Slug: the local project name and category slug. The slug defaults to
ComfyUINodeBuilder/so you can append your own pack folder, for exampleComfyUINodeBuilder/PromptTools. Deploy mirrors the active pack intocustom_nodes/ComfyUINodeBuilder/<pack-folder>.
- Choose a template in the Nodes panel and click New.
- Add inputs, outputs, widgets, and Return UI items from Add to Node.
- Edit node metadata and the visual contract in the preview.
- Edit the generated Python file in the code workspace.
- Open the AI panel to create, revise, validate, test, or explain the node.
- Review generated files such as
__init__.py,requirements.txt,install.py, and runtime UI scripts. - Export a ZIP or click Deploy & Restart to write into ComfyUI.
- Refresh ComfyUI and test the generated node in a workflow.
The README is the overview. The wiki is the detailed guide:
- Wiki index
- Getting started
- Using the workbench
- Creating nodes
- Deploying to ComfyUI
- Project structure
- Troubleshooting
npm run dev # Start Vite and the loopback helper server
npm run server # Start only the helper server
npm run mcp # Start the local MCP server integration
npm run build # Type-check and build the frontend
npm run preview # Preview the production build
npm run test # Run the Vitest suite- The app is a local development workbench, not a hosted service.
- Direct deploy requires file-system access to a local ComfyUI installation.
- Restart requests depend on ComfyUI Extension Manager support.
- Browser project registry and AI settings are stored in local storage.
- Sensitive helper actions are restricted to loopback origins and loopback ComfyUI URLs where applicable.
- Existing custom node packs without
builder.project.jsonare not automatically converted into editable builder projects.
src/ Vue app, stores, components, and client-side pack builders
server/ Express helper server for local file, AI, terminal, and ComfyUI actions
tests/ Vitest unit and component tests
docs/wiki/ User-facing guide and wiki pages
docs/superpowers/ Design and implementation notes for this repo
site/ Static GitHub Pages landing page
public/ Static browser assets
ComfyUI Node Builder 是一个本地 GUI 应用，用来轻松构建 ComfyUI 自定义节点和完整节点包。它把可视化节点接口、生成的 ComfyUI 节点包文件、Python 源码编辑、AI 辅助开发、本地检查、ZIP 导出和托管部署放在同一个工作台里。
如果你想把一个节点想法快速变成可运行的 ComfyUI 自定义节点，而不想每次都手写同一套样板结构，这个项目适合你。
它适合：
- 想提高编辑、测试、部署效率的 ComfyUI 自定义节点作者
- 正在原型设计图像、latent、mask、音频、文本或工具节点的工作流作者
- 想减少样板代码、但仍然要直接控制 Python 源码的开发者
- 希望使用 AI 辅助开发，同时保留可见文件变更和验证步骤的原型开发者
- 需要构建带依赖、安装脚本、Return UI 或前端运行时显示的复杂节点包的人
ComfyUI 自定义节点很强大，但开发循环很重复：
- 定义 Python 类和
INPUT_TYPES - 同步返回类型、返回名称、映射和分类
- 添加控件、自定义端口和 Return UI
- 管理
requirements.txt、install.py和共享文件 - 每次部署后重启 ComfyUI
- 在生成文件漂移时排查问题
这个项目把这些步骤整理成一个结构化的本地工作台。
AI Builder 直接内置在工作台里，不是一个独立的聊天页面。
它可以帮助你：
- 根据提示创建节点包和节点
- 修改选中节点代码和生成文件
- 解释验证错误或终端错误
- 验证生成的节点包
- 在节点包工作目录中运行终端检查
- 在开启 Allow deploy 后请求部署
- 通过已安装的
codexCLI 使用本地 Codex 计划，或使用 OpenAI、OpenRouter、OpenAI-compatible、Anthropic、Gemini、Ollama 等模型提供商
部署是有门禁的。AI 可以请求部署动作，但只有你明确开启允许部署后，界面才会发出部署操作。
- 可视化节点接口： 添加输入、输出、控件、自定义类型、Return UI、分类和元数据。
- 完整 Python 编辑： 在 Monaco 工作区中直接编辑生成的节点 Python。
- 生成节点包文件： 检查节点文件、
__init__.py、运行时 UI 脚本、自定义渲染器、requirements、安装脚本和共享文件。 - 托管部署： 导出 ZIP，或把 builder 管理的节点包写入
<ComfyUI>/custom_nodes/ComfyUINodeBuilder/<pack-folder>/。 - 可回载元数据： 通过
builder.project.json让 builder 创建的节点包可以再次加载回应用。 - 本地 helper server： 文件写入、AI 调用、终端命令、依赖安装和 ComfyUI 重启请求都走 loopback。
要求：
- 推荐 Node.js 22 或更新版本。
- 如果要直接部署或加载节点包，需要本地 ComfyUI 目录。
- 如果希望部署后自动请求重启，推荐安装 ComfyUI Extension Manager。
安装并运行：
git clone https://github.com/caoool/comfyui-node-canvas
cd comfyui-node-canvas
npm install
npm run dev开发脚本会启动：
- Vite 前端，通常是
http://localhost:5173 - helper server：
http://127.0.0.1:3001，开发环境中通过/helper代理
打开 Vite 地址，然后从状态栏进入 Settings，配置：
- ComfyUI URL： 通常是
http://127.0.0.1:8188 - ComfyUI Install Path： 包含 ComfyUI
custom_nodes目录的文件夹 - Pack Name / Pack Slug： 本地项目名和分类标识。Slug 默认是
ComfyUINodeBuilder/，可以在后面追加自己的节点包目录，例如ComfyUINodeBuilder/PromptTools。部署会同步到custom_nodes/ComfyUINodeBuilder/<pack-folder>。
- 在 Nodes 面板选择模板，然后点击 New。
- 从 Add to Node 添加输入、输出、控件和 Return UI。
- 在预览区编辑节点元数据和可视化接口。
- 在代码工作区编辑生成的 Python 文件。
- 打开 AI 面板，让 AI 创建、修改、验证、测试或解释节点。
- 检查
__init__.py、requirements.txt、install.py和运行时 UI 脚本等生成文件。 - 导出 ZIP，或点击 Deploy & Restart 写入 ComfyUI。
- 刷新 ComfyUI，在工作流中测试生成的节点。
README 是概览，详细手册在 wiki：
- Wiki index
- Getting started
- Using the workbench
- Creating nodes
- Deploying to ComfyUI
- Project structure
- Troubleshooting
如果它帮你减少了 ComfyUI 自定义节点样板代码，欢迎给仓库一个 Star。Issue、PR、真实使用场景和 AI 辅助节点开发需求也都很有价值。