Visit ComfyUI Online for ready-to-use ComfyUI environment
ComfyUI_NetDist enables running ComfyUI workflows across multiple local GPUs or networked machines, featuring nodes for remote images and local remote control.
ComfyUI_NetDist is an extension designed to enhance the capabilities of ComfyUI by allowing you to run workflows on multiple local GPUs or networked machines. This extension is particularly useful for AI artists who need to leverage the power of multiple GPUs to speed up their image generation processes or distribute the workload across different machines. By using ComfyUI_NetDist, you can achieve faster rendering times and more efficient use of your hardware resources.
At its core, ComfyUI_NetDist works by distributing the image generation tasks across multiple GPUs or machines. Think of it as a team of artists working together on a large mural. Instead of one artist doing all the work, each artist (or GPU) is assigned a portion of the mural. This collaborative effort results in a faster and more efficient completion of the artwork.
Here's a simplified breakdown of how it works:
This feature allows you to control multiple GPUs on a single machine or across different machines. You can set different ports and CUDA devices to manage the GPUs effectively.
Ideal for users with two GPUs or two separate PCs. This setup requires minimal configuration and allows you to set local/remote batch sizes and trigger nodes based on specific conditions.
This setup scales the dual-GPU example by connecting more nodes together. It helps in distributing the workload across multiple machines, although it may require some adjustments to avoid duplicate images.
For more complex setups involving more than two GPUs, this feature allows for easier per-batch overrides and setting default batch sizes. It also supports using workflow JSONs as inputs, making it highly customizable.
The LoadImageUrl
and SaveImageUrl
nodes allow you to load and save images via URLs. This is useful for integrating with web services or remote storage solutions.
This feature enables the seamless passing of latent data between nodes. You can save latents as .npy
files and load them from other instances, facilitating complex workflows that require intermediate data sharing.
Currently, ComfyUI_NetDist does not include different models. It focuses on distributing tasks across multiple GPUs or machines to optimize performance and efficiency.
--listen
to your launch arguments and set the correct remote IP. Check your IP using ipconfig
on Windows or ip a
on Linux.For additional resources, tutorials, and community support, you can visit the following links:
© Copyright 2024 RunComfy. All Rights Reserved.