ComfyUI-Workflow-Finder Finds ComfyUI Workflows by Describing Them

By vramkickedin | | 2 min read
a sleek magnifying glass with a frosted glass lens focuses on a tiny abstract network of interconnected nodes.

The ComfyUI-Workflow-Finder is a new desktop tool that lets you search through your local collection of ComfyUI workflow files using plain English descriptions. It indexes workflows by their node structure, so you can find what you need without remembering file names. The app also includes a live web search feature to discover workflows from online sources like YouTube, CivitAI, and Reddit.

Developer Gregowahoo created this tool after collecting over 1,300 workflows and struggling to locate them quickly. He built it as a standalone Python application that runs locally, so you keep full control of your data. The project addresses a common pain point for ComfyUI users who accumulate hundreds of workflows and need a smarter way to navigate them.

Node pack filter and graph preview

Key Features
  • Semantic search using natural language descriptions.
  • Optional Claude mode for advanced AI search.
  • Sort results by creation or modification date.
  • Preview node graph with zoom and pan.
  • Color-coded Bezier wires show data types.
  • Filter workflows by installed custom node packages.
  • Live web search of YouTube, CivitAI, and more.
  • Automatic download link detection for results.

ComfyUI users with large local workflow collections can finally search by function, not just filename. Professionals and agencies who rely on ComfyUI for daily work will appreciate the node pack filter that ensures compatibility before opening a file. Privacy-conscious users can run the fast search mode completely offline, with no data leaving their machine.

Early development and local-first design

The tool is distributed as a single Python script, requiring only Python 3.10+ and Tkinter to run on Windows, macOS, or Linux. The AI-powered search and Find in the Wild feature need an Anthropic API key, but fast keyword search works entirely offline. The project has no formal release yet, but the source code is available under the MIT license for anyone to modify or contribute.

"Search your local workflows by describing what you want ("generate video from an image", "face swap with LoRA") — not just by filename " — Source: Reddit

Project Links