Open 0X-JonMichaelGalindo opened 2 weeks ago
low Varm unfriendly
High vram friendly. I suggest maybe making it as an option perhaps?
I went ahead and added my changes to an existing PR I created, otherwise the PR would have probably failed https://github.com/AIFSH/OmniGen-ComfyUI/pull/13. Thanks for the inspiration @0X-JonMichaelGalindo
Attach the OmniGen pipeline to the node instance as an attribute to avoid unloading the model from VRAM after each generation completes. Is this the right way to do this? It works in my local Comfy test, but I am not sure.