Type | |
Stats | 567 |
Reviews | (52) |
Published | Apr 2, 2023 |
Base Model | |
Training | Steps: 3,000 |
Trigger Words | sks woman |
Hash | AutoV2 58816300C2 |
This is my first LyCORIS.
I believe that the LyCORIS quality is superior to that of a regular LORA.
In general, my technique has not changed. I still train a dreambooth model and then extract the result. Previously I've been extracting it as a LORA, now it will be a LyCORIS.
I still keep all the trained Dreambooths but I feel like it is no longer necessary to upload them as the LyCORIS quality is "good enough".
Samples were done on various models. You need to test the weight on each model as different checkpoint actually prefers different weights.
RealisticVision tends to work with 1.0 (probably because that was the base of the dreambooth training) but others need between 0.8 and 1.4 usually.
Realistic models will give us realistic outputs, stylized models will give us stylized results :)