Supergirl with the white collar. I trained this LoRA to learn how to build LoRAs on a Linux system running an AMD 9070 XT.
Thanks to the Fizzgig project for making it far easier than ever to train LoRAs locally. This is probably my most successful attempt yet for creating a LoRA on ROCm on Linux.
Some additional prompts may be needed to keep her design consistent. Her wristbands are currently the biggest issue with the generation. Manually prompting seems to help somewhat. If you see issues, try:
her right wristband (viewer's left) is red, her left wristband (viewer's right) is blue, solid red wristband, blue wristband with yellow stripeor
her right wristband (viewer's right) is red, her left wristband (viewer's left) is blue, solid red wristband, blue wristband with yellow stripeThere are other character inconsistencies, but they can be rendered inconsequential with additional positive or negative prompts. Hopefully I can iron out the inconsistencies as I learn the LoRA training system better.
Description
The good enough version. Training crashed on epoch 78 on a 2000 epochs target.
