Type | |
Stats | 2,667 17,820 |
Reviews | (352) |
Published | Aug 7, 2023 |
Base Model | |
Training | Steps: 2,000 Epochs: 8 |
Trigger Words | AngelicStyle |
Hash | AutoV2 D1D2AA4E4B |
My first SDXL model! SDXL is really forgiving to train (with the correct settings!) but it does take a LOT of VRAM ðŸ˜! It's possible on mid-tier cards though, and Google Colab/Runpod! If you feel like you can't participate in Civitai's SDXL Training Contest, check out our Training Overview!
LoRA works well between 0.8-1.0. Use trigger word AngelicStyle to boost the output. All training images were with the subject facing the front, with symmetrical wings, so you may have trouble getting anything else! Full disclosure; Some faces and hands in-painted in the sample images, and one was img2img from SD 1.5.
The LoRA also has a nice effect when added to random prompts - lots of backlighting and epic aura!
If you're trying to get started with SDXL, check out our Civitai getting started guide, and our SDXL training overview for tips and settings!