Remember to use the trigger word leechaeyoung. You can use LORA weight 1 and Hires fix with latent bicubic antialiased, 20 steps, 0.4-0.6 denoising strength, and 1.25x upscale, BUT I think hires fix isn't really necessary, for example, the classroom pictures are actually the before hires fix versions. I've used <a target="_blank" rel="ugc" href="https://civarchive.com/models/6424/chilloutmix">ChilloutMix</a>, <a target="_blank" rel="ugc" href="https://civarchive.com/models/17449/level4">Level4</a>, and <a target="_blank" rel="ugc" href="https://civarchive.com/models/9052/lofi">LOFI</a>.This is my favorite LORA I've made in terms of likeness. The only issue is that the dataset is pretty biased with closeups of Chaeyoung looking left. She just has a lot of pictures showing off her jawline (understandable). You'll have to weight "looking at viewer, from front, upper body, cowboy shot" accordingly. Also delaying the face composition with "[:(detailed face, perfect eyes, glossy lips:1.2):0.2]" helps too with avoiding close-ups.

Files with SHA256: ba7064440869b5394472f6201929e2aa052aac172730eae1a3465f77e0b1278c