ComfyUI-QING
A powerful all-in-one ComfyUI extension with comprehensive professional nodes for AI models, image/mask/SVG processing, video synthesis, and workflow management. Features modular architecture with node alignment tool and multi-platform API support.
Nodes (79)
Flip any boolean — the NOT gate ComfyUI forgot to ship
True becomes 1, False becomes 0 — and suddenly your graph math works
Turn a hex code into a color name your prompt actually wants
Zip up your whole ComfyUI setup before the next upgrade breaks it
Restore a ComfyUI environment from a zip — without nuking what you have
Turn a category list into embedding vectors — no API, but it needs one extra install
Load an SVG in ComfyUI without watching LoadImage choke on it
Gemini image editing against the gateway of your choice
Inpaint or edit through any OpenAI-compatible image endpoint
Point a chat node at any OpenAI-compatible LLM — local vLLM included
Your own endpoint for vision — any OpenAI-compatible VLM, one key
DeepSeek inside the graph, pick your hosting platform
ByteDance's Doubao vision — strong on UI and document understanding
A counter that ticks every run — your batch loop finally knows where it is
Tell Gemini to change the background to a blue sky, get the image back
Gemini's vision in your graph — with a thinking knob for 2.5 models
GLM chat in your graph — Zhipu's model, your choice of host
Ask a GLM vision model what's actually in your image
Did that node actually produce anything? This answers in three flavors
Hold 99 images in memory, then dump them in one batch
Crop by exact pixel coordinates — and get a mask for free
Crop by percentage — the fastest way to trim a composition
Shave exact pixels off each edge — and keep the mask
Stop guessing what's inside your image tensor
Mirror an image (or a whole batch) in one click
Image ↔ mask conversion, two channels at a time
Finally, a mask preview that's actually readable
Image Rotation
Resize the image and its mask together — or they drift apart
Turn a raster into vector paths — without leaving ComfyUI
The two-second node that unblocks your logic wiring
Turn a Number Into Text in ComfyUI (Int to String)
Drop Moonshot's Kimi into your graph — long context, real work done
Ask an image questions — Kimi reads it for you
Peek inside any wire in the graph — without ripping the flow apart
Pull one item out of a list — with a safety net for bad indexes
Feed images to ComfyUI's API as base64
Skip the upload, just paste links
Load a .txt or .csv into the graph — straight from disk
Get SVG markup into the graph without touching a browser
Four-channel terminal logging that doesn't break your wires
Eight blend modes to turn several masks into one region
Know exactly what your mask covers — before you burn a sampler pass
Grow your selection before you inpaint
Fill the holes in a mask — the fix for every donut-shaped selection
Know whether you actually have a mask before you feed one downstream
Scale a mask to a target size without touching the image it came from
One mask, five objects — split it into a mask per component
Stitch model-00001-of-00005.safetensors shards back into one file
Collect a run's output files into one archive and get a download link
Stop hand-typing file paths into a dozen string widgets
Accumulate images run after run into one growing batch
Pull one image out of a batch at a time, with a loop counter
A four-way switch for any data type, driven by booleans
Recover the original filename of the image in your pipeline
A free, local 'is this image any good?' check — no VLM required
Stamp your own prompt, seed, and model onto an output image
When the first API call fails, hand the graph the backup result
Wildcard-style prompt variation without the wildcard syntax
Fill {var1}..{var8} placeholders in a prompt template
Save an image under a name you control, to a folder you choose
The fastest way to see what's actually flowing through a wire
Chat with Qwen inside your workflow — without running a single local parameter
Ask a vision model what's in the image, right inside the graph
Pick a random file from a folder — reference, LoRA, or prompt — as a path
The debug node that just shows you the raw value
The exact strings that count as 'true' here
StringToInt — convert a text value into an integer ComfyUI can use
Get that SVG out of the graph and onto disk, with a preview you can trust
Bringing your vectorized line art back into the graph
Three string equality checks in one node, and the empty-string trap inside it
ComfyUI's printf — with a timestamp and a level
Four ways to cut a string, with the pieces on separate ports
One template, five slots, and a debug report on what filled them
Free text into a fixed category list, via embeddings — and the setup it quietly demands
The JSON you feed your vLLM-Omni server when the main node ran out of widgets
Image editing against your local vLLM server — and when it quietly switches endpoints
Text-to-image through your vLLM-Omni server, two endpoints in one node
A chat call to any OpenAI-compatible server, from inside the graph
ComfyUI-QING · AI创作全能工具箱
让AI创作更简单,让工作流更智能。
ComfyUI-QING 是一个全面的 ComfyUI 扩展包,提供 79 个精心设计的专业节点,覆盖 AI 大模型接入、图像处理、遮罩工程、SVG 矢量图形、数据流控制、调试与配置管理六大领域。全面采用 ComfyUI Nodes 2.0 纯原生架构,启动极速、运行稳定。
🎯 为创作者而生 — 设计师、动画师、AI 艺术家、开发者,都能在这里找到提升效率的利器
🚀 Nodes 2.0 纯原生 — 100% 静态化适配,毫秒级扫描加载,完美兼容最新 ComfyUI 前端
🤖 AI 深度集成 — 内置 9 大主流 AI 平台,支持语言对话、视觉理解、图像编辑
🌍 国际化支持 — 内置中文/英文界面,随 ComfyUI 语言设置自动切换
🧩 径向菜单系统 — Pizza Slice 风格,25+ 插件,10 大对齐工具,大幅提升画布效率
⚡ Nodes 2.0 纯原生架构
全面采用 ComfyUI Nodes 2.0 纯原生规范,拥有极致的启动速度和完美的向后兼容性:
- 极速加载 — 基于文件签名缓存的智能节点发现,毫秒级扫描
- 精准呈现 — 原生 Schema 定义,完美适配最新 ComfyUI 前端特性
- 无缝兼容 — 自动兼容旧版节点,现有工作流无需修改
🎨 径向菜单插件系统
引入优雅的 Pizza Slice 径向菜单,完全融入 ComfyUI 原生主题:
- 插件化架构 — 内置 25+ 专业插件,涵盖工作流管理、快速创建节点和实用工具
- 10 大对齐工具 — 支持节点上/下/左/右对齐、水平/垂直居中、等距分布、拉伸匹配,极大提升画布整理效率
- 多面板与快捷键 — 支持最多 3 个独立配置的径向菜单面板,可自定义快捷键防冲突
- 历史管理 — 内置操作历史追踪,支持撤销/重做
📊 节点总览
| 分类 | 节点数 | 核心能力 | |------|:-----:|---------| | 🤖 AI 调用 | 19 | 覆盖 DeepSeek、GLM、Kimi、Qwen、Doubao、Gemini 等 9 大平台,支持语言对话、视觉理解、图像编辑、自定义 API 端点 | | 🖼️ 图像处理 | 16 | 旋转/翻转/缩放/裁剪、Base64/URL 加载、图像缓存池、质量评估、图像转 SVG、遮罩互转 | | 📊 数据工具 | 25 | 语义向量匹配、类型转换(6向)、条件路由、批量处理、提示词模板/随机化、文本分割/拼接、路径组装 | | 🎭 遮罩工程 | 6 | 智能拆分、8 模式混合、方向扩张、内部孔洞填充、可视化预览、数据统计判断 | | 🔧 配置与调试 | 10 | 工作流导入/导出、多级调试面板、日志打印、模型分片合并、文件打包输出、元数据写入 | | 📐 SVG 处理 | 3 | 本地加载/保存、自动格式校验、SVG → PNG/JPG 栅格化、位图 → SVG 矢量化 |
✨ 核心能力详解
🤖 AI 大模型接入(19 个节点)
支持多平台、多模态的统一 AI 调用框架:
| 平台 | 节点 | 能力 |
|------|------|------|
| 智谱AI (GLM) | GLMLanguageAPI GLMVisionAPI | 语言对话、视觉理解 |
| 月之暗面 (Kimi) | KimiLanguageAPI KimiVisionAPI | 长文本分析、图像问答 |
| 阿里百炼 (Qwen) | QwenLanguageAPI QwenVisionAPI | 文本生成、视觉推理 |
| 火山引擎 | DoubaoVisionAPI | 豆包视觉理解 |
| 硅基流动 | DeepSeekLanguageAPI | DeepSeek 语言对话 |
| Google AI | GeminiVisionAPI GeminiEditAPI | Gemini 视觉问答与图像编辑 |
| VLLM 全通 | VLLMT2TAPI VLLMT2IAPI VLLMI2IAPI | 文→文、文→图、图→图,支持 OpenAI 兼容端点 |
| 自定义 URL | CustomURLLanguageAPI CustomURLVisionAPI CustomURLImageEditAPI | 对接任意 OpenAI 兼容 API |
| 故障转移 | QING_ModelFallback | 模型调用失败时自动切换备用方案 |
特色能力:
- 内置多轮对话缓存与上下文管理
- 统一的流式输出与错误处理
- 自定义 URL 节点可对接任何 OpenAI-compatible 服务
- VLLM 全通节点支持高级参数微调
🖼️ 图像处理(16 个节点)
- 基础变换 —
ImageRotation(旋转)、ImageFlipping(翻转)、ImageScaling(缩放)、MaskScale(遮罩缩放) - 精确裁剪 —
ImageCropByOrigin(坐标裁剪)、ImageCropByPixels(像素裁剪)、ImageCropByPercent(百分比裁剪) - 多源加载 —
CustomLoadImageWithFormat(本地多格式)、LoadImageFromURL(远程 URL)、LoadImageFromBase64(Base64 解码) - 图像缓存 —
ImageCache智能缓存池,最大支持 99 张轮转 - 质量评估 —
QING_ImageQualityScore图像质量评分 - 格式互转 —
ImageMaskConverter(图像↔遮罩)、ImageToSVG(位图→矢量) - 命名工具 —
QING_GetImageName(提取名称)、QING_SaveImage(保存)
📊 数据工具(25 个节点)
- 语义匹配 —
VectorMatcher(类别向量匹配)、ConvertDataToVectors(数据→向量)、ColorJudge(颜色判断) - 类型转换 —
DataConverter系列:Int↔String、String↔Bool、Bool↔Int、Bool 翻转,6 向无缝转换 - 流程控制 —
QING_ConditionRouter(条件路由分支)、HasValue(值存在判断)、DynamicIntegerIncrement(动态自增) - 批量处理 —
QING_BatchForEach(逐项处理)、QING_BatchCollect(结果收集) - 提示词工具 —
QING_PromptRandomizer(随机提示词)、QING_PromptTemplate(模板填充)、TextTemplateConcat(模板拼接) - 文本处理 —
TextCompare(多文本对比)、TextSplitter(分割)、LoadPathText(文件加载)、PathAssembler(路径组装) - 数据工具 —
ImageDataAnalyzer/MaskDataAnalyzer(数据统计分析)、ListItemGetter(列表索引取值)、RandomDirectoryItem(随机目录项)
🎭 遮罩工程(6 个节点)
- MaskSplitter — 智能拆分复杂多元素遮罩,支持文字/图形保护的智能合并
- MaskBlend — 高级混合,支持 8 种混合模式
- MaskExpansion — 方向性扩张,精细控制各方向扩展量
- MaskFill — 智能内部孔洞填充
- MaskJudgment — 遮罩数据统计与条件判断(面积、数量等)
- ImageMaskPreview — 图像与遮罩的混合实时预览
🔧 配置与调试(10 个节点)
- ConfigExport / ConfigImport — 一键导出/导入完整 ComfyUI 工作流及 Python 依赖环境,支持跨端快速恢复
- LetMeSee / ShowMePure — 任意数据查看器,右键节点快速接入专属调试面板
- LogPrinter / QuickLogPrinter / TextLogPrinter — 多级日志输出
- MergeShardedModel — 模型分片合并
- OutputFilePack — 文件打包输出
- QING_MetadataWriter — 图片元数据写入
📐 SVG 处理(3 个节点)
- LoadSVG — 本地 SVG 文件加载与格式校验
- SVGSaver — SVG 内容保存到文件
- SVGToImage — SVG → PNG/JPG 高质量栅格化(推荐配合 cairosvg)
📥 安装指南
1. 源码安装
进入 ComfyUI 的 custom_nodes 目录:
cd ComfyUI/custom_nodes
git clone https://github.com/GAOSHI-QING/ComfyUI-QING.git
2. 依赖安装
推荐使用自动化安装脚本:
cd ComfyUI-QING
python install_dependencies.py
国内用户使用镜像加速安装:
python install_dependencies.py --mirror --auto
更多选项:
| 参数 | 说明 |
|------|------|
| --mirror | 使用国内镜像源(阿里云)加速 |
| --auto | 自动安装全部依赖,无需手动确认 |
| --skip-optional | 跳过可选依赖(svglib、reportlab) |
| --check-only | 仅检查依赖状态,不安装 |
| --silent | 静默模式,适合脚本自动化 |
或手动安装:
pip install -r requirements.txt
3. 环境要求
- Python ≥ 3.9
- ComfyUI 最新版本
- 推荐安装
cairosvg以获得最佳 SVG 渲染效果 - 网络支持 — AI API 节点需要网络连接
🔑 API 密钥配置
重启 ComfyUI 后,在顶部菜单栏选择 设置 (Settings) → QING → API配置,可视化填入各平台的 API Key:
- 所有配置本地热保存,重启不丢失
- 节点调用时自动拉取,无需在工作流中暴露密钥
- 支持 9 大平台独立配置
🌍 国际化
内置中英文界面翻译,随 ComfyUI 语言设置自动切换。翻译文件位于 locales/ 目录,欢迎贡献其他语言。
📁 项目结构
ComfyUI-QING/
├── nodes/
│ ├── api/ # AI 大模型接入(19节点)
│ ├── image/ # 图像处理(16节点)
│ ├── data/ # 数据工具(25节点)
│ ├── mask/ # 遮罩工程(6节点)
│ ├── utils/ # 配置与调试(10节点)
│ ├── svg/ # SVG 处理(3节点)
│ └── video/ # 视频处理(开发中)
├── js/ # 前端 JS(径向菜单、调试面板、设置同步等)
├── locales/ # 国际化翻译(zh / en)
├── example_workflows/ # 示例工作流
├── data/ # 数据配置模板
├── scripts/ # 构建与工具脚本
├── install_dependencies.py # 依赖安装脚本
├── requirements.txt # Python 依赖
└── pyproject.toml # 项目元数据与构建配置
🤝 参与共建
欢迎提交 Issues 反馈问题,或通过 PR 贡献新功能!无论是节点优化、新平台适配、格式扩展还是文档完善,你的参与都能让这个工具更强大。
<br>
<a id="english"></a>
ComfyUI-QING · All-in-One Creative Toolkit for ComfyUI
中文 | Nodes | Install | API Config
Making AI creation simpler, workflows smarter.
ComfyUI-QING is a comprehensive ComfyUI extension delivering 79 meticulously crafted nodes across 6 domains: AI model integration, image processing, mask engineering, SVG pipelines, data flow control, and debugging & configuration. Fully built on the ComfyUI Nodes 2.0 pure native architecture for lightning-fast startup and rock-solid stability.
🎯 Built for creators — designers, animators, AI artists, and developers alike
🚀 100% Nodes 2.0 native — no legacy adapters, millisecond-level scanning, perfect UI compatibility
🤖 9 AI platforms — language, vision, and image editing via a unified framework
🌍 i18n ready — built-in Chinese & English UI, auto-switching with ComfyUI locale settings
🧩 Radial menu — Pizza Slice style, 25+ plugins, 10 alignment tools for massive canvas productivity
📊 Node Overview
| Category | Nodes | Highlights | |----------|:-----:|------------| | 🤖 AI APIs | 19 | DeepSeek, GLM, Kimi, Qwen, Doubao, Gemini + Custom URL & VLLM Omni for OpenAI-compatible endpoints | | 🖼️ Image | 16 | Transform, crop, multi-source loading (local/URL/Base64), cache pool, quality scoring, raster↔vector | | 📊 Data | 25 | Semantic vector matching, 6-way type conversion, condition routing, batch processing, prompt tools | | 🎭 Mask | 6 | Intelligent splitting, 8 blend modes, directional expansion, hole filling, visual preview | | 🔧 Utils | 10 | Workflow import/export, multi-level debug panels, logging, model shard merging, metadata writing | | 📐 SVG | 3 | Load, save, bidirectional SVG↔PNG/JPG conversion |
✨ Key Features
🤖 AI Model Integration
- DeepSeek, GLM (Zhipu), Kimi (Moonshot), Qwen (Alibaba), Doubao (ByteDance), Gemini (Google) — language, vision, and image editing
- VLLM Omni — text→text, text→image, image→image via any OpenAI-compatible endpoint with advanced parameter tuning
- Custom URL nodes — connect any OpenAI-compatible API for language, vision, or image editing
- Model fallback — automatic failover on API errors
- Multi-turn conversation memory with context management
🖼️ Image Processing
- Rotation, flipping, scaling (multiple interpolation methods), mask scaling
- Crop by origin coordinates, pixel dimensions, or percentage
- Load from local path, remote URL, or Base64 string
- Smart image cache pool (up to 99 images in rotation)
- Image quality scoring
- Raster→SVG vectorization, image↔mask conversion
📊 Data & Flow Control
- VectorMatcher — semantic category matching via embedding vectors
- DataConverter series — 6-way conversion between Int, String, and Bool
- ConditionRouter — flow branching based on text comparison
- BatchForEach / BatchCollect — batch processing pipeline
- PromptRandomizer / PromptTemplate — dynamic prompt generation
- Type-agnostic tools: text splitting, path assembly, list access, random directory item, dynamic increment
🎭 Mask Engineering
- MaskSplitter — split complex multi-element masks with text/shape protection
- MaskBlend — 8 blend modes for advanced compositing
- MaskExpansion — directional expansion with per-side fine-tuning
- MaskFill — intelligent internal hole filling
- MaskJudgment — statistical analysis and conditional checks on mask data
- ImageMaskPreview — real-time blended preview
🔧 Configuration & Debugging
- ConfigExport / ConfigImport — full workflow & dependency snapshot for cross-machine recovery
- LetMeSee / ShowMePure — attach debug viewers to any node via right-click context menu
- MergeShardedModel — merge model shards into a single file
- OutputFilePack — file packaging for output management
- QING_MetadataWriter — write metadata into generated images
📥 Installation
cd ComfyUI/custom_nodes
git clone https://github.com/GAOSHI-QING/ComfyUI-QING.git
cd ComfyUI-QING
python install_dependencies.py
Users in China — use the mirror flag for faster downloads:
python install_dependencies.py --mirror --auto
Requirements: Python ≥ 3.9, ComfyUI latest.
cairosvgrecommended for best SVG quality.
🔑 API Key Setup
After restarting ComfyUI, go to Settings → QING → API Configuration to enter your API keys. All keys are saved locally and auto-injected at runtime — never exposed in your workflow graphs.
🤝 Contributing
Issues and PRs are welcome! Whether it's node optimization, new platform support, format extensions, or documentation improvements — your contribution makes this toolkit better.
- 🐛 Report an Issue
- 📖 Documentation
- ⭐ Star the project to support ongoing development
ComfyUI-QING · Unleash your creativity. 🚀