Replies: 4 comments
-
Fundamentally, ComfyUI was not designed for multi-user use, and AI image generation utilizes the GPU fully. |
Beta Was this translation helpful? Give feedback.
-
... however, a concurrent "parallelism" would be much appreciated. Here's a use case for a single-houshold or even single-user multi-workflow scenario (i.e., when security/privacy isn't an issue):
Yes, it's a very rudimentary "parallelism", but even it would be a massive improvement in terms of multitasking. |
Beta Was this translation helpful? Give feedback.
-
Do you need this for the parallel development of workflows, or just for using existing ones? If you only need it for using existing workflows and not for development, there are solutions available that aim to do that and use ComfyUI under the hood, such as StableSwarmUI or Visionatrix |
Beta Was this translation helpful? Give feedback.
-
Thanks for the suggestion. I know of StableSwarm, but it's not a real solution for an underlying problem. It's a workaround at best. In reality, you almost never "develop workflows" and "use them" as two separate processes (the latter is an activity for complete beginners and, as a result, for youtube bloggers on the other end - as an effective tool for self-promotion and TEASING what can be done with ComfyUI to someone who has never saw any node editor before). And since ComfyUI has no support for subgraphs, at least splitting one enormously giant spaghetti mess into smaller, a bit more manageable messes is kinda the obvious way to go for any kind of complex networks. Until we get proper subgraphs, of course. So no, I don't think a GUI wrapper like Swarm is a true solution for the issue. The true one is finally getting subgraphs in the core... and I believe getting multiple tabs to work concurrently is a necessary step to get there. |
Beta Was this translation helpful? Give feedback.
-
I have implemented a ComfyUI server on a Mac machine. Currently, users connect to the same address and use the server simultaneously. The problem is that all the process (image generation) do not occur simultaneously but are queued as if there were a single user. This causes delays.
Is this the correct approach to use a server on a network for multiple users? Is there a solution?
Thanks!
Beta Was this translation helpful? Give feedback.
All reactions