ComfyUI Extension: ComfyUI-Replicate

Authored by replicate

Created

Updated

165 stars

Run a/Replicate models in ComfyUI.

Custom Nodes (0)

    README

    comfyui-replicate

    Custom nodes for running Replicate models in ComfyUI.

    Take a look at the example workflows and the supported Replicate models to get started.

    example-screenshot

    Set your Replicate API token before running

    Make sure you set your REPLICATE_API_TOKEN in your environment. Get your API tokens here, we recommend creating a new one:

    https://replicate.com/account/api-tokens

    To pass in your API token when running ComfyUI you could do:

    On MacOS or Linux:

    export REPLICATE_API_TOKEN="r8_************"; python main.py
    

    On Windows:

    set REPLICATE_API_TOKEN="r8_************"; python main.py
    

    Direct installation

    cd ComfyUI/custom-nodes
    git clone https://github.com/replicate/comfyui-replicate
    cd comfyui-replicate
    pip install -r requirements.txt
    

    Supported Replicate models

    View the supported_models.json to see which models are packaged by default.

    Update Replicate models

    Simply run ./import_schemas.py to update all model nodes. The latest version of a model is used by default.

    Add more models

    Only models that return simple text or image outputs are currently supported. If a model returns audio, video, JSON objects or a combination of outputs, the node will not work as expected.

    If you want to add more models, you can:

    • add the model to supported_models.json (for example, fofr/consistent-character)
    • run ./import_schemas.py, this will update all schemas and import your new model
    • restart ComfyUI
    • use the model in workflow, it’ll have the title ‘Replicate [model author/model name]’

    Roadmap

    Things to investigate and add to this custom node package:

    • support for more types of Replicate model (audio and video first)
    • showing logs, prediction status and progress (via tqdm)

    Contributing

    If you add models that others would find useful, feel free to raise PRs.