ComfyUI-Inspire-Pack - Backend Cache Nodes

When ComfyUI executes a workflow with a different node configuration, all caches are cleared, preventing data sharing between workflows. This results in the need to start from the beginning when switching workflows.

In this video, I introduce backend cache nodes that allow data sharing when switching between multiple workflows.

[Najdmie]

Will definitely try this. My workflow is too large that I frequently have to pan and zoom. I don't like switching tab because data is not shared. This solves this problem.

[xueqinghuang]

This is a great nodes at workflow, I use it immediately

[roktecha]

This is really great - I've been using it to replace noodles in the same workflow. Any downsides you can think of to using it this way?
UPDATE: I think I figured out the draw of using it intra-workflow -- it can throw off node execution sequencing -- still I think I'll keep at it for a while.

[drltdata]

You can download from ComfyUI from here:

And ComfyUI Inspire Pack extension from here:

And ComfyUI Impact Pack extension from here:

And ComfyUI Workflow Component extension from here:


Workflow:

[LeKhang]

Awesome thank you very much for sharing. Is it possible to do the opposite too? My biggest workflow has 30-40 loras and when I generate 3-4 times the CPU Mem (Not GPU) is full (RTX 3090). When I just start ComfyUI the Mem usage is 28% then 70% >> 85% >> 99% >> Crashed. When that happens I have to kill ComfyUI, load libtcmalloc and then start it again to reduce Memory usage to 28%. On github they said it's ComfyUI's memory leak issue so I'm searching for a way Freeing up CPU memory after each generation automatically.

[riggitywrckd]

Very interesting, a couple of times I've wanted to drag drop images directly from a web search into comfyui running on another tab in the same browser. Does anyone know if this functionality exists? It would be very straightforward compared to right click, save file, over to comfy for the loading. I think you can drag drop from the file browser but not direct inside the browser. Does anyone have that working? I really like this latent cashing idea, could jump around and use the same latent like when you want to start animatediff off of something it could be worked into the graph to start it off a latent generated by another graph without having to mess around so much.