日本語

ComfyUI-Jumper

Designed to integrate local ComfyUI with remote ComfyUI(ex: Runpod, Vast.ai)<br /> Pass string and images between multiple ComfyUI via socket.<br /> YouTube Demo<br />

On Remote On Local

Install

cd ComfyUI/custom_nodes/
git clone https://github.com/myonmu0/ComfyUI-Jumper

Usage

1. Rent one server, or start 2 ComfyUI on local.<br />

2. Open a port on remote, MUST use a secure channel like SSH Tunnel.

ssh root@REMOTE_IP -p REMOTE_PORT -NL 8282:localhost:8282

• Default port are 8282, however you can change this in nodes.py DEFAULT_ADDR val, or manually change on node.
• If you will integrate local with local, skip this.<br />

3. Start local and remote ComfyUI. As tutorial, load the examples/local.json to local ComfyUI, examples/remote.json to remote ComfyUI and run both to see if works.

• Accessing remote ComfyUI with SSH Tunnel are secure, easy and simple way that works on all cloud provider.
• Accessing remote ComfyUI with browser's private/secret mode is a cleaner option because starts with 0 browser data, and ends with 0 data.<br />

⚠️ Known Issue: Job cancel button don't work when this node are running, in such case restart ComfyUI.

Details

Select all node "mode" to "Client" on local, and "Server" on remote.

1 Jumper node chain on 1 workflow.

The node "order" determine the order of sending/receiving data, and must be chain correctly on both side. Don't forget to add at least 1 node on the end of Jumper node chain(like in the example)

| send_as | Size | CPU | About | | --------- | ------------- | --------------- | ------------------------------------------------- | | png | Small | Moderate | Convert tensor to png before sending. _{Progress bar: 1 = Conversion, 2 = Sending} | | zlib | Large | Moderate | Compress tensor before sending. | | raw | Very Large | Very Low | Raw tensor are send. |