ComfyUI Extension: ComfyUI-Transformers-Pipeline

Authored by mediocreatmybest

Created

Updated

3 stars

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

Additional ComfyUI nodes to utilise the Transformers pipeline in a simple and modular way.

README

ComfyUI Transformers Pipeline

Overview

A collection of additional ComfyUI nodes for utilizing the Hugging Face Transformers pipeline in a modular and efficient manner.

Installation

Can be installed through the ComfyUI manager or ComfyUI cli

Nodes

Primary Nodes

  • Model Loader: Loads transformer models.
  • Caption Generator: Generates image captions.
  • Batch Processor: Simple node to specify and handle lists to batch processing.
  • Export Captions: Simple node to save text output to files based on original file input names.

Utility Nodes

  • Debug Node: Allows connecting list nodes to a text output to view node outputs
  • Debug Model Node: Allows connecting the model loader node to a text output to view node outputs and check configuration
  • EXIF Metadata Extractor: Extracts camera metadata from images.
  • Task List Node: Provides preset transformer tasks.

TODO

Short Term

  • [ ] Add tooltips nodes.
  • [ ] Move pipeline nodes into its own folders (that’s our primary focus really) with non-pipeline tasks separated.
  • [ ] fix logging to possibly Python’s logging module.
  • [ ] Add additional example nodes

Long Term

  • [x] Register in ComfyUI registry
  • [ ] look into multi-pass captions or outputs
  • [ ] add additional pipeline tasks
  • [ ] add multimodal tasks to the mix
  • [ ] Add additional output nodes
  • [ ] Allow setting different file extensions for output

Contributing

Pull requests are welcome. As well as Ideas, suggestions, anything at all. What works, what doesn’t, etc.

License

Please see license file.

Run ComfyUI workflows without the setup

No installs, no CUDA version roulette, no GPU sitting idle on your bill. Bring a workflow and run it in the browser.

Learn more