Facial ???????? for Qwen Image. Trained on ~250 photos all of completely different women (no "multiple appearances" in the dataset whasoever), and captioned using hand-checked natural language descriptions from jailbroken Gemini 2.5 Pro. No real particular "trigger word" other than cum obviously. Prompting with natural language descriptions of the appearance you want tends to work best.

I went out of my way to make the dataset as diverse in terms of both ethnicity and age ranges from early-to-mid-20s up to around mid-50s-ish as I could, so this Lora should be fairly flexible in terms of being able to get fairly specific "looks".

Recommended inference strength is around 0.6 to 0.8 or so, but feel free to experiment as you see fit. All sample images were generated with Euler Beta @ CFG 4.

Generating at resolutions in the 1 megapixel range (1024x1024, 832x1216, 1216x832, 896x1152, 1152x896, 768x1344, 1344x768, and so on) and then upscaling from there with a second denoise pass if desired will likely give you better overall results than generating with only one pass directly at Qwen's "default" set of higher ~1.75 megapixel resolutions.