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

Feature: Better LoRA management #1507

Closed
moonwhaler opened this issue Dec 22, 2024 · 5 comments
Closed

Feature: Better LoRA management #1507

moonwhaler opened this issue Dec 22, 2024 · 5 comments

Comments

@moonwhaler
Copy link
Contributor

moonwhaler commented Dec 22, 2024

This might be a bigger one: I work with LoRAs a lot (maybe others do too). I have collected hundreds of LoRAs over the last months.

The current implementation of the LoRA management is straightforward, but very limited in terms of UX.

I would love to have some sort of browser (with preview images grabbed (manually) from e.g. Civitai.com and some sort of automatic trigger word implementation.

I use Civitai.com Browser+ for Forge to collect this information (which places a text file and a preview image next to the LoRA (or model checkpoint) with the same filename).

This might not be high on any list, but it would make life a lot easier.

Thank you.

@rktvr
Copy link

rktvr commented Jan 1, 2025

agreed

@Acly
Copy link
Owner

Acly commented Jan 23, 2025

Note that trigger words are supported in general:
Image

Picking them up from text files: this is a bit complicated because the plugin doesn't scan folders (and they might be on some remote server). Lora files are reported by ComfyUI. So first support for the text files and exposing them would have to be implemented there (can be done as a custom node).

Preview image: that would be a lot of work and not very useful for most people without also implementing a way to grab those images from civitai. I'm probably too rigorous with deleting crappy Lora to ever feel motivated enough tbh

@moonwhaler
Copy link
Contributor Author

Thanks for the info. I already looked at the code and saw you're refreshing the LoRA list when starting the plugin and save it locally. One could theoretical make an API call to an meta data provider at refresh time, which would certainly impact startup time. As I said, it's more of a nice to have feature for me.

What I noticed is that refreshing the LoRA or model list after the plugin started (using the settings dialog) takes a huge amount of time when comparing the same process at startup time. Sometimes it takes up to 60-100 seconds for a complete refresh, when having 500+ LoRAs on the server side (locally).

Maybe I open an issue.

@Acly
Copy link
Owner

Acly commented Jan 24, 2025

Might also be related to checkpoints if you have a lot of those (they are always refreshed together, even though there are separate buttons).

@moonwhaler
Copy link
Contributor Author

moonwhaler commented Jan 27, 2025

After a bit more digging I started writing a little "CivitAI" scraper myself, which is also able to sync trigger/activation words directly to the "loras.json".

"metadata": { "lora_triggers": "word1, word2" }

So yeah. Getting this included into Krita AI Difusion seems overkill.

@moonwhaler moonwhaler closed this as not planned Won't fix, can't repro, duplicate, stale Jan 31, 2025
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

3 participants