About this model:

This model has a greater resemblance to the features of Zendaya's face. In addition, it also has greater overall aesthetic quality and the ability to adhere to the prompt. The process that was applied to do so was as follows: 1-A model with BS=1 was trained with images in 768x768 resolution. The images were rescaled with Supir. This training was with a 1e-4 learning rate. 2- The original images were taken (those rescaled with Supir), and given as input (img2img) to flux, using this initial lora, with a noise of 0.5. 3-A model was trained with BS=4, and 512x512 resolution with the images rescaled with Flux and the 768x768 resolution lora, LR= 3e-4. The result is a lora with greater aesthetic quality, and a greater facial resemblance is also maintained. In my opinion, the other Lora that I have published about Zendaya is better, because I give more importance to the physical resemblance of the entire body, and less to the face. The lighting on this model is incredible, it does fantastic things, and even more so if you use an image as input.

Recommended parameters:

Sampler: DPMM_2S_ANCESTRAL

Scheduler: Beta

Steps:20-30.

Note: most demo images are pending moderation approval. So you don't have many examples to look at, I recommend you download it and look at it for yourself.