Comfyui Workflow Generator

ComfyUI工作流生成器。从自然语言描述生成ComfyUI工作流。基于ComfyGPT研究架构，经历生成→验证→构建三阶段管道。触发词：ComfyUI工作流、生成工作流、ComfyUI自动化。

Security Scan

The package and runtime instructions are internally consistent with a local ComfyUI workflow-generator: it scans the local ComfyUI node registry, loads local LLM models (GGUF/HuggingFace/llama-cpp), and builds workflow JSON — nothing requests unrelated credentials or hidden network endpoints in the provided artifacts.

Purpose & Capability

Name/description (generate ComfyUI workflows from NL) match the code and README: node generator, validator, and builder modules, node cataloging, model loaders, and model manager are all needed for that functionality.

Instruction Scope

SKILL.md and README instruct the user to run an Update Node Catalog which scans the local ComfyUI nodes and catalog files; the code reads local ComfyUI modules, model files, and writes catalog/TinyDB files. This is expected for a node that validates compatibility, but it does mean the skill will read local ComfyUI state and model directories (no extra env vars or unrelated system files are referenced in the provided snippets).

Install Mechanism

There is no platform-level install spec (instruction-only at registry level) but full source is bundled. README instructs cloning and pip installing requirements; it requires llama-cpp-python for GGUF models which may need local compilation. This is proportionate but raises normal build/runtime friction (native build, GPU drivers).

Credentials

The skill requests no environment variables or external credentials. It operates on local ComfyUI paths and local models; configuration resolves paths under ComfyUI custom_nodes and ComfyUI/models/LLM which is coherent for its purpose.

Persistence & Privilege

Flags show no forced always: true and auto-invocation is allowed (the platform default). The code registers a local folder/type for LLM models and writes catalog files, which is expected behavior and limited to the custom node's scope.

Assessment

This implementation appears to do what it claims, but take these precautions before installing: 1) Review requirements.txt and install the node in an isolated Python environment or container since it depends on torch, sentence-transformers, llama-cpp-python (which may require native compilation). 2) The "Update Node Catalog" step scans your local ComfyUI installation and writes catalog/TinyDB files — that's expected but means the node reads local ComfyUI modules and metadata. 3) Models are expected to be local (GGUF or HuggingFace); check whether you will download models from HuggingFace or provide them locally — automatic downloads can occur only if you run the provided loader code and have network access. 4) If you are unsure, run the node in a disposable VM/container and inspect the created files (catalog/, ComfyUI/models/LLM/) and network activity during initial runs. 5) If you need deeper assurance, review utils/model_loaders/huggingface_loader.py and any code that fetches remote models to confirm there are no unexpected remote endpoints or credentials used.

基于自然语言描述生成ComfyUI工作流的AI工具。

核心功能

• 自然语言 → ComfyUI工作流（JSON）
• 三阶段管道：Generator（生成）→ Validator（验证）→ Builder（构建）
• 支持 SDXL / Flux / Video 等多种节点类型
• 模块化设计，可在每个阶段介入检查

触发词

「ComfyUI工作流」「生成工作流」「帮我写ComfyUI流程」

使用限制

• 依赖ComfyUI本地环境
• 无法识别训练截止日期后发布的新节点
• 建议在使用前检查工作流的节点兼容性

版本

• Skill 版本: 1.0.0