Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Ability to load multiple LoRA for SD Image Generation #1181

Closed
guxanan opened this issue Oct 21, 2024 · 5 comments
Closed

Ability to load multiple LoRA for SD Image Generation #1181

guxanan opened this issue Oct 21, 2024 · 5 comments

Comments

@guxanan
Copy link

guxanan commented Oct 21, 2024

Describe the Issue
There are a variety of LoRA for different purposes, such as better image generation with much fewer needed steps like SDXL-Lightning, different art styles, scene compositions, characters.

Being able to load only one is very limiting, even more so for those with weaker hardware who want to try various models in combination with SDXL-Lightning with other LoRA, so it would be amazing to have this feature available.

Usecase

  • Using a standard SDXL/Pony model, loading the SDXL-Lightning LoRA for image generation with fewer steps, and then loading a style LoRA, character LoRA, etc.

Additional Information:

  • gpu: AMD Radeon RX 590 Series (radeonsi, polaris10, LLVM 18.1.8, DRM 3.59)
  • cpu: AMD Ryzen 5 3600 6-Core Processor
  • koboldcpp: 1.76
  • system: "x86_64-linux"
  • host os: Linux 6.11.2-xanmod1, NixOS, 24.11 (Vicuna), 24.11.20241014.a3c0b3b
  • multi-user?: yes
  • sandbox: yes
  • version: nix-env (Nix) 2.24.9
  • nixpkgs: /nix/store/xnjw9gmfmpppdj6bxpw6cfkspc3h6xwl-source
@guxanan guxanan changed the title Ability to load multiple lora for SD Image Generation Ability to load multiple LoRA for SD Image Generation Oct 21, 2024
@LostRuins
Copy link
Owner

Have you considered merging some of them together?

@guxanan
Copy link
Author

guxanan commented Oct 21, 2024

Have you considered merging some of them together?

If koboldcpp provides a script for it, I didn't know. I would like it if you could kindly pointed me to some resources for that as I'm still quite new to this.

Using such a method wouldn't be a great solution though. For example, quickly changing between weights/strength for different LoRA wouldn't be possible and you'd have to make multiple merges with different weights and keep them all on disk, which would increase space usage substantially.

I'm not aware of how speedy such a process would be on weaker hardware either, and my hardware barely qualifies as low-end at this point.

I'm sorry if this is a big task to ask for, I would love for this feature to be implemented, but I understand if you don't want to do it.

@LostRuins
Copy link
Owner

Yeah unfortunately for the moment I think only a single lora is supported.

@guxanan
Copy link
Author

guxanan commented Oct 21, 2024

Unfortunate but understandable, should this issue be closed or remain open?

@LostRuins
Copy link
Owner

I'll close it for now, may be revisited in future

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants