This is the "0.5" version of this LoRa. It was captioned way too much, making likeliness a bit hard to achieve. You can use it in ComfyUI at strength 1. I picked all captions provided during training and ran them through an LLM to find the patterns and elaborate a new prompt that can activate the Lora properly. <pre><code>" ohwx, yu lexuan, the image is a high-resolution photograph of a young East Asian man dressed in traditional attire. He has a serene expression, with long, dark hair neatly tied back and adorned with a white headpiece. His fair complexion and delicate features are accentuated by the soft, even lighting.He is wearing a flowing blue robe with wide, elegant sleeves. The fabric has a subtle sheen, suggesting it is made of fine silk. A white inner layer peeks from beneath the robe, adding contrast to the ensemble. The waist is cinched with a light blue sash, featuring a small decorative ornament.The background is a stark white, ensuring that all focus remains on the subject and his attire. The image captures the subject in a poised stance, evoking a sense of historical elegance and cultural refinement. The overall composition highlights the timeless beauty of traditional clothing, emphasizing the subject's dignified presence."</code></pre>It can be refined even further, but some expressions do need to remain as they are key captions to the model:ohwx, yu lexuan - activation words"white headpiece" is strongly tied to his face, not including it can cause problems.This lora is very experimental and was trained using only images. I found no benefits using videos, even though hundreds of small clips I can use.

Files with SHA256: d67447de56d2fc35b9796ae95b885584a6cd3bfce3e885dbd3de9eddf9b7b78f