A LoRA trained on images hands without a body, they're holding, grabbing, etc. The only needed trigger is disembodied hand. I do recommend prompting the action the hand is doing for better results, check the images above for examples.