Weight of 1.0 works well most of the time.
I am not asking for coffee, but if you like my work, a kind review is always appreciated.
Description
No trigger words needed
FAQ
Comments (7)
amazing, such great detail, coherence, and a lot of fun
This is amazingly unique, awesome work!
Ok secrets spill now:
How did you reduce the dim and still get it to be this strong LOL.
HOW DID YOU REDUCE THE DIM on an SDXL lora and still get it under 50mb :P
The SDXL model is 4x the size (roughly). So you only need 1/4 the dim to have the same size. I have had much better luck with sdxl with smaller dims. 8 or 4 linear dims does amazing things on sdxl. More is not always better
@Ostris oh I'm aware but 32dim vs 64 ended up being worse .. it's a crapshoot. I can do 8dim 1.5 Lora and in fine ... It must be more dataset and learning rate
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.