The prompting will need more details to get a good image. See my sample image prompts.
The data images where all photoshopped, the original images where of woman with wings and I edited the arms out. You will get some images, with arms and some even look like a bad edit of removed arms. I found some checkpoints work better than others, you will have to experiment.
use trigger word - "sirnwobrd"
Also use "photo of " helps also.
I had some images of birds with woman's head in the training and those show up once in a while. Some have folded wings with wings spread wide. Rendering more images in a batch gives you a chance for good images.