multigpu_diffusion_comfyui

ComfyUI nodes to run multigpu_diffusion.

(Basically, a ComfyUI front-end to Diffusers, with support for multi-GPU environments.)

Notes:

• Requires Diffusers-format checkpoints/models.
• Does not (and will never) use native sampling.
• Windows and macOS are not (and probably never will be) supported.
• This node uses psutil/subprocess to manage the host scripts. You can find the calls in modules/host_manager.py.

Setup:

1. cd to custom_nodes/multigpu_diffusion_comfyui.
2. (Review and) run setup.sh.

Debugging:

• If you run into issues, you can manually curl to the hosts (e.g., curl localhost:{port}/close to manually shut down the host).

Known Issues:

• Running balanced host will create device errors, use another host for now.

Test Environment:

• ComfyUI 0.12.3, front-end 1.16.9
• 4x Nvidia Tesla T4
• Python 3.14.4
• Torch 2.14.0