Type | |
Stats | 396 |
Reviews | (61) |
Published | Mar 29, 2023 |
Base Model | |
Trigger Words | filianore |
Hash | AutoV2 82F7EFB8E5 |
update v25:
v25 version is a Lycoris (kohya locon), you may need install extension which supports the format to use.
The v25 version focuses on full body portrait/profile drawing, suggest weight 0.7-1.0 and CFG 3.
When generating full body picture, use 512x768 as the default resolution and use hires-fix/multi-diffusion/SD upscale to fix the face. It may need 1280x1920 to get a clear face.
If you had installed the LoRA block weight extension, you could use the following block weights to get better(not so overfit) results at CFG7 for realistic style.
<lora:Filianore_v2:1:1,0.5,0.5,0.7,0.7,1.2,0.5,1,1,0.5,1,1,1,0.5,1,1,1,1,0.5,1,0.75,0.75,1.2,0.5,1,1>
v1:
Training a Lora of Filianore is very difficult because she has very few picture. I put a lot of effort into this and the result is not too bad.
Suggest weight 0.9-1.0, CFG 3-7. The lower the CFG the closer the results tend to painting.