nodes/SeanScripts-ComfyUI-Unload-Model
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Update README.md

Browse Source
This commit is contained in:
SeanScripts
2024-09-02 17:36:15 -07:00
parent 8a5e6a4489
commit 348ee89c3c
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
README.md
+5 -1
View File
@@ -2,7 +2,11 @@
For unloading a model or all models, using the memory management that is already present in ComfyUI. Copied from https://github.com/willblaschko/ComfyUI-Unload-Models but without the unnecessary extra stuff.
How to install: Clone this repo into the `custom_nodes` folder, then restart ComfyUI.
Includes two nodes: Unload Model and Unload All Models. These are used as passthrough nodes, so that you can unload one or all models at a specific step in your workflow.
The main use of this is to unload the CLIP model after getting the embedding for the prompt, because it's just a waste of VRAM to keep it loaded while sampling, and VRAM is at a premium with Flux. For example, if you use the Q4_K_S version of Flux dev, with unloading the CLIP model, only about 10 GB of VRAM is used when generating at 1024 x 1024, which means that users with 12 or 16 GB GPUs might be able to run the model without going into shared memory, which could be a huge speedup. Please let me know if you test it and see this benefit. This could also be useful if you want to free up VRAM so you can keep a local LLM loaded for prompt generation.
How to install: Clone this repo into the `custom_nodes` folder, then restart the ComfyUI server.
How to use: Add in the middle of a workflow to unload a model at that step. Use any value for the `value` field and the model you want to unload for the `model` field, then route the output of the node to wherever you would have routed the input `value`.