roundglassesvB2used the 48 images from vB1 but added masks and trained at 8 dim.roundglassesvB1also trained on PonyRealism_v21VAE. 48 images selected from the images that trained vs1, only using the ones that had the largest glasses. 19 anime, 29 realistic.I get best results using score_9, score_8_up, score_7_up, score_6_up in the prompt along with score_5_up, score_4_up add in the negative. This version can be used with the previous version. So far I have only done lower values with the previous version when combining.roundglassesvs1Trained on <a target="_blank" rel="ugc" href="https://civarchive.com/models/372465/pony-realism">PonyRealism_v21VAE</a> with 143 images, 87 anime/cartoon style images, 44 realistic style, and 13 images that were semirealistic. All images were labelled with round glasses. The anime/cartoon were also labelled with source_anime. The realistic were also labelled with source_photo, realistic. I didn't label the semirealistic with any extra labels. None of the real images had any mature themes but in the anime there were a handful of explicit images so to avoid any explicit images in your generations I recommend using source_explicit in the negative.Keywords: round glasses, source_photo, source_anime, realistic,*when wanting realistic images use source_photo and/or realistic in positive and source_anime in negative.*when wanting anime/cartoon styles use source_anime in positive and source_photo in negative.The image examples are only using this LoRAUsed these prompts in all images:score_9, score_8_up, score_7_up, score_6_up BREAK1girl,round glasses&lt;lora:RoundGlassesVs1:0.75&gt;negative:score_5_up, score_4_up, source_explicit,*beside 1girl, I used the name of the character.

Round Glasses

Trained on same data set as vB1 but I added masks for each image and trained at 8 dim.

Files with SHA256: ddcfd7f61879765103f193cb0e0df45610d38e6169ecda83191582ad957c0398