I recommend using "simple background" and/or "yellow background" or any other defined color. You can also add "sphere" and "cables" to positive or negative tags depending on what you want to prompt. There's a lot of images tagged as "extra arms" but they're so varied the tag doesn't work reliably.
For sample prompts see the uploaded images and the zip with training tags. Note that you have to change the LoRA name if you copy my sample prompts, the filename was misspelt as "ub1imo" when they were generated.

There's a lot of images with variable number of limbs and non-human themes which can make image generation a bit chaotic. I recommend generating a large batch at lower resolution first to find a good base image and then using the "Variation seed" option in Automatic1111 to refine it further. You can also use ControlNet poses to help improve results.

MISC
Trained with 120 on NAI at 512 resolution.
All sample images are raw txt2img, cherrypicked, no inpainting or helper embeddings used.
The training was done on a small subset of the artists images, as such the output of this LoRA is not an accurate representation neither in quality or style of the source material. Go support the artist ub1mo if you want the good stuff.