Lora's (and TI's) can be heavily influenced by the checkpoint you're using. One way to get the best likeness, when the accuracy and/or likeness is of the utmost importance, is to use a checkpoint that used the same base model as the base model used for training the Lora. (There are exceptions - i.e. when the checkpoint has been trained for a specific look for faces. Note the difference in RyanBlend v5 vs RyanBlend v6 directly below it. Both have a base model of SD15 but v6 has been trained for a specific face look.)
Checkpoints and Lora's usually have their base model listed in the side panel on the right of the page. If you're going for the best possible likeness of your images, I encourage you to double check the base model of your checkpoint and Lora. You'll get the best likeness if the base models match.
Every sample image has the exact same settings and prompts. The only difference is the checkpoint used for image generation.
POSITIVE PROMPTS: High detail RAW color Photo of a gorgeous woman, (face closeup:1.2) portrait, smirk|smiling, brunette, [(freckles:0.7)|(moles:0.6)], eos, skin_pores, 8k high definition, high detail eyes, highest quality, skin texture, masterpiece, best quality, 8k, perfect lighting, cinematic lighting, (sharp focus:1.2), HDR, beauty, ultra-detailed, amazing
NEGATIVE PROMPTS: (3d:1.6), (render:1.6), (3dcg:1.6), (deformed:1.6), different, bad, (ugly:1.3), draft, drawing, duplicate, error, (extra:1.1), failure, fake, floating, grainy, noisy, gross, illustration, corruption, (corrupt:1.5), (irregular:1.3), (compressed:1.2), (compression:1.2), (artifacts:1.3), (lopsided:1.5), low, (lowres:1.3), (malformed:1.3), misshapen, monochrome, mosaic, multiple, mutated, mutation, mutilated, old, oversaturated, surreal, twisted, unappealing, uneven, unnatural, weird, worst, hairy, pimples, acne, tattoos, rolls, folds, bad_pictures, easynegative, ng_deepnegative_v1_75t, Unspeakable-Horrors-64v, blonde