In my experience training models, I've noticed that for some uncommon or very specific concepts, it's better to separate them in the captioning so that the AI knows exactly how they look, or else in this case, it doesn't learn the eyes quite well. I'm not sure if it's the correct method, but personally, it has worked for me.

Hello! Although I trained the model specifically to include the shackles, if you want to generate without them, it's possible but a bit tricky. You need to decrease the weights of Lora and the word 'Briarlol',Also, putting 'hands in shackles' and 'shackles' in negatives works for me. I just uploaded some images. I hope they serve as an example for you. Obviously, it will lose some characteristics of her clothing, but hopefully, you can achieve a good image :D