My Links: discord, twitter/ig
Subscribe for FBB images @ https://patreon.com/fitCorderAI
Use this with ComfyUI.
Combine then and facedetail with this https://civarchive.com/models/104503/comfy-fc-suite-fcxl
I’m working on testing different datasets and seeing how things work. Results will vary for a while. For now I hope you enjoy testing on the cutting edge. Results will improve as we as a community learns more.
Description
FAQ
Comments (13)
Interesting!
I haven't played with XL yet - is that the base model that's making all the preview images have women with Giraffe Neck torsos?
These images were significantly better anatomy then those without the Lora.
yeah... its the base model doing that
girls faces acceptable even great but anatomy...not yet
How certain are we that LORAs and fine-tunes trained on 0.9 will be valid for use on 1.0?
I was informed they "should" work but we don't really know. Eitherway its really good to start getting used to XL as XL is the future with 1024x1024 being the base resolution.
Nice to see the first releases for sdxl! I tried training it myself, but can't get decent results .. it's my first time using kohya ( used Stable Tuner before for 1.5) Epoch 1 looks promising but then it tends to get more and more muddy instead of more refined. No idea what's going wrong ... Could you share your trainig config? That would be really aweseome !
for now i'm using the preset in bmaltis for sdxl - lokr
@fitCorder you mean the sdxl-24GB-lokr_v1 ? Have you changed the Learning Rate or LR Scheduler? Or keeping everything at the preset Values?
@solocon Prodigy optimizer handles the learning rate