A somewhat successful attempt at making a LoRA for better hands. Still struggles a bit with prompt adherence and "spaghetti fingers", but seems stable enough to publish.
The most repetitive part of image generation for me has always been the hands. In Illustrious and NoobAI, this is painfully obvious, as the hands tend to look the exact same, often with thick fingers which are hard to remove by prompting.
In my testing, the LoRA can be quite extreme. I'd suggest toning the weight down to about 0.7-0-8 if you're planning to capture more than just hands.
Trained on NoobAI v1.1 (e-pred)
If you're reading this, check out my new software for dataset sorting and captioning! I'd love some feedback if anyone ends up trying it!
Feel free to ask questions, whether it would be about the software above, the training parameters, or anything else!