This lora is a test using a fully synthetic dataset. All training images were generated using Albedo Base XL since illustrious itself cannot provide the images the way I expected them to come. You don't necessarily need a lora for this, you can generate a generic Korean male using illustrious, then use any SDXL model as a refiner for a more realistic look. Yet I wanted a lora for it for better control. This model works best with WAI. 2DN and other Illustrious checkpoints I have access to do not perform so well. No real photo of a BTS member was used to train this, so if you find some inconsistency with the real members, this lora was not made to be true to life. The model has no activation token. You can use these tags:<pre><code>white suit, bts, korean, solo, realistic, brown hair, formal, suit, short hair</code></pre>

Files with SHA256: 2f103be7d03c1a8e3bb18ef65e9f111b674b0278b38548e540d53f6adfc550e9