lora of a fictional woman named madisonput owl in the negative prompt if it starts generating owlspurpose of this is to catalog progress as it improves in qualityV001: i generated at 1216x1216 then scaled down to 1024 to attempt to remove gridlines and seems to have improved the quality greatly from madison, next try will use an upscaler. I also trained 50 images instead of 25. By adding a name for the character into the prompts (sara) it seems to have upped consistency of the face greatly allowing for more images. This was also trained with horizontal flip augmentation.Wan 2.2 14b: You can mix this with any lora at 1+ strength and turn the woman in the video into this character. Depending on the lora you're using the faced may change and you can try to increase the strength of the character or decreasing the strength of the lora you're using to compensate but it's entirely dependent on the quality of the lora. No keyword necessary, just say "Woman" somewhere in the prompt and make sure you have both the high and low lora.

Files with SHA256: 5adb816264bbd2c4fd856a2d6b0d985b9345b230da7bed6670274bd9a414275b