Select a subgraph
Choose nodes and use Cloud Offload selection. The box holds GPU, provider, timeout, and warm-runner settings.
01 / COMPUTE & INFRASTRUCTURE
Your graph.
A bigger GPU.
Run the demanding part of a ComfyUI workflow on a cloud GPU. Your nodes stay visible and editable, with remote progress, previews, and errors reported in place.
ONE GRAPH. TWO PLACES.Workflow
LOCAL
Load Imageimage input
Cloud Offload remote GPU
Load Modelrunner model
Sampler + Decodegeneration
LOCAL
Save Imageresult output
Keep your nodes. Move the compute.01 → 02 → 03
HOW IT WORKS
Move the compute, keep the workflow.
Review the rental
Free preflight recommends a GPU and shows price, estimated cost and time ranges, cache coverage, and uncertainty before paid work begins.
Watch results arrive
Progress and previews return to the original nodes. The Cloud Jobs panel tracks transfers, execution, cancellation, and resource billing.
TWO REPOSITORIES, ONE SYSTEM
A small client.
A separate coordinator.
The node pack speaks HTTP to the service. Provisioning and provider credentials live in the coordinator.
ComfyUI-Cloud-Offload
The editable box, workflow bridges, live node feedback, rental confirmation, and job panel.
Browse the node packcloud-offload
The service that manages provisioning, execution, queues, and remote workers. RunPod is the default provider; Vast.ai is an alternative.
Browse the serviceGET STARTED
Give your graph somewhere to run.
Install and configure the coordinator first. The node pack needs a running service; it is not a standalone GPU provider.
1. Set up the coordinator
Follow the service README for installation, provider credentials, and a worker image. Then start the service:
python -m cloud_offload serve --host 127.0.0.12. Install the node pack
comfy node install cloud-offloadOr clone it into ComfyUI/custom_nodes and restart ComfyUI.
git clone https://github.com/jethac/ComfyUI-Cloud-Offload.git