Here in the community, many train Lora elements with low parameters, but this Lora element was trained with rank 64, which absorbs a greater amount of information, tends to be stronger, and delivers better results. Small Lora elements will have lower quality, with a less polished appearance and noisy images.

I use Learning Rate=5-e5, Network Rank Dim=64, Network Alpha=84, Epoch=8, max step=1200 settings because that's when I get the best quality at LORA weight=1. If I were to use a lower quality for training, then LORA would have its full effect at weight=1.5, but then LOAR would really settle on all other objects in the scene. I use step=1200 because ideally, with two or three times more steps, the quality would be adequate with low LORA settings, but it costs a lot of energy and I don't have enough money to power my RTX-4090 video card responsibly and I have to cool my room in the summer...so it's not worth it. My LORAs are of high quality because I use special captions that allow you to achieve different types of scene results in the prompts during rendering, but they all have one drawback: the size of the LORA is around 500 MB.

Look at the other LORAs, you can see that the prompts of the images use the words of the captions, this allows you to use LORA as if you were rendering with a real 3D scene program. The words and sentences of the captions train LORA much better, they also contain the smaller elements of the scene, so the embeddings in the latent matrix are much better controlled, they approach the given FLUX1d model embeddings much more precisely. Sorry if I'm not trying to explain the LORA size-quality problem at a basic level, but I don't know how familiar you are with the mathematics and processes of AI rendering.