Haven't tried, I think it should also works,

internally, lora weights are added to UNet, neg_prompt will be encoded to unconditional embeddings and merge with positive prompt, basically the same to prompt but will have negative impacts on the UNet denoising. So, it should works in theory.