What is comfyui-mesh?

ComfyUI-Mesh is a tool for splitting large AI image and video generation models across two separate graphics cards (GPUs), either on the same computer or on two different machines connected over a local network. Normally, running a big model like FLUX.2 or LTX 2.3 requires a single very powerful GPU with enough memory to hold the entire model at once. This project solves that problem by dividing the model into a front half and a back half, each running on a different GPU. The clever part is how it handles the data transfer between the two cards. Modern Nvidia GPUs contain dedicated video compression chips called NVENC that normally sit idle during AI work. ComfyUI-Mesh uses those chips to compress the activation data (the intermediate results being passed between the two model halves) before sending it over the wire, much like compressing a video stream. This cuts the transfer size by three to ten times, making a regular gigabit home network fast enough for real-time use. The result is that FLUX.2 Klein 9B can generate a 1024-pixel image in roughly 4.4 seconds split across two cards over ethernet. The system has two parts: Icarus, a custom node installed inside ComfyUI (the popular AI image generation interface) on your main machine, and Daedalus, a server that runs on the second machine. You would use this if you have two Nvidia GPUs that individually do not have enough memory to run a large model on their own, or if you want to pool a friend's GPU over a VPN connection. It currently supports FLUX.2 Dev, FLUX.2 Klein 9B, and LTX 2.3 video models, with more architectures planned.