ComfyUI Extension: ComfyUI-Offload-Models

Authored by lokinou

Created

Updated

1 stars

Custom nodes to offload and rapatriate models from cpu.

Custom Nodes (0)

    README

    ComfyUI Offload models

    Manually offload and reload models between devices (cpu, cuda ...). Useful when restricted on VRAM.

    Usage

    Place the nodes around the model you want to execute, connecting an input through (i.e. generated image) that serves as trigger.

    • "Offload Model" after processing, allowing to send the model to cpu
    • "Recall Model" sending back the model to cuda

    Requires a bit of control on the workflow execution, it can be more convenient than deleting models and their cache.

    Getting started

    Install

    Via ComfyUI

    Search for "comfyui-offload-models" in the available node, or paste the github repo link to install from it.

    Manual Install

    # Go to the custom nodes
cd ./ComfyUI/custom_nodes
# Install the repo
git clone https://github.com/lokinou/comfyui-offload-models.git

    Updates

    • 1.1.0 : Added enable button, device button and error behavior. Now catches unsupported models and throw error messages or ignores.
    • 1.0.0 : now working as intended, many fixes with faulty code, improved node names, individualized modelpatcher and model which may be on different devices at once,
    • 0.0.2 : Both offload the modelpatcher and the referenced model, otherwise it doesn't free VRAM
    • 0.0.1 : First version uploaded, with an Experimental flag

    Examples

    TODO