Type | |
Stats | 249 4,611 462 |
Reviews | (48) |
Published | Sep 28, 2023 |
Base Model | |
Training | Steps: 750 Epochs: 1 |
Usage Tips | Clip Skip: 2 |
Trigger Words | Kindred Lamb White fur |
Hash | AutoV2 22B4DAC82E |
This is my first try at creating a Lora, so this might be far from perfect.
Despite it, some result are pretty goods (and others pretty bad).
The point was to port SD 1.5 Lora to SDXL. The whole process is pretty basic, but botched due to my inexperience.
This Lora was trained on ProtoVision XL v0.5.2.0 using 10 images generated in 512*768 (and upscaled x2 with hires using Ultrasharp), with SD 1.5 model Yiffymix (https://civitai.com/models/3671/yiffymix) and a Kindred Lora for this SD version (https://civitai.com/models/79462/kindred-or-lamb-or-league-of-legends-lora)
The result was then fed to Kohya_ss with a 1024*1024 target.
The prompt would look like this :
"1girl, white fur, white skin, kindred, lamb"
with another addition :
"black mask, blue eyes, glowing eyes" or "umasked, blue white eyes"
If you have any experience with creating Lora, and/or tips, tutorial and the like to optimize the resullts (and the training time, considering an RTX 4060 16GB), I would be most gratefull, since I'd love to experiment more in the future.