Description
Originally trained as a 128/128 dim/alpha lora i decided to retrain it as a locon, the parameters for the locon were the following 768 resolution, batch size 3, 5e-4 unet lr, 5e-5 text encoder lr, 16/8 dim/alpha and 16/8 conv dim/alpha. I honestly dont know what version is best since both work well but maybe the original lora was a bit undercooked, regardless just try both and decide for yourself.
For the locon version you will need to either install additional networks extension or this other one https://github.com/KohakuBlueleaf/a1111-sd-webui-locon
FAQ
Comments (5)
the locon should have its own model page vs the lora so that people can use the categories correctly
i saw other people doing this to keep it more concise.
@kar the other people are wrong too ._.
@aisignin what is even the difference? Why provide both formats? Which should I use and how to decide?
@MetaSamsara before some version you needed an extra extension for locon which was a hassle for some people rn they're implemented on the base webui version so usually locon are better for art styles.
Details
Available On (2 platforms)
Same model published on other platforms. May have additional downloads or version variants.