For best results, I recommend using this LoRA at 0.6–0.7 weight for back-view generations, and around 1.0 weight for front-view generations.

Please note that this LoRA sometimes affects hand quality. To help improve the results, I strongly recommend adding the following to your prompt:

hands wrapped in tape,

I also found that when using angle-related tags such as from side, from front, or from above at the same time — especially for three-quarter style views — the arms are likely to show perspective errors.

I plan to release a V2 after I improve the issues mentioned above.