robertvoy/ComfyUI-Distributed

What it solves

ComfyUI-Distributed allows users to scale their AI image and video generation by distributing workloads across multiple GPUs and machines. It solves the problem of being limited by a single GPU's processing power, enabling faster batch generation and accelerated upscaling workflows.

How it works

The extension adds specialized nodes to ComfyUI that orchestrate the distribution of tasks. It uses a master-worker architecture where a master node coordinates the process and dispatches tasks to workers. These workers can be local GPUs on the same machine, remote GPUs on other computers in a network, or cloud-hosted GPUs (e.g., via Runpod).

Who it’s for

Artists and developers using ComfyUI who have access to multiple GPUs—either locally or in the cloud—and want to increase their output volume or speed up high-resolution upscaling.

Highlights

• Parallel Workflow Processing: Generate multiple images or videos simultaneously using different seeds across multiple GPUs.
• Distributed Upscaling: Accelerates the Ultimate SD Upscale by splitting image or video tiles across multiple workers.
• Flexible Worker Types: Supports local, remote, and cloud-based GPU workers.
• Orchestrator-Only Mode: The master machine can act solely as a coordinator without performing any rendering itself.
• Developer API: Provides a programmatic endpoint to control the distributed cluster without using the browser UI.