Another retrain. Did a lot of cleaning up and had to source some extra images for her secondary outfit. Ended up doing some Img2img to some settei images as they were too crappy and low resolution which rather inflated the training time. Regardless one cute Cornelia makes the effort worth it I guess. I stopped mid baking to test <a rel="ugc" href="https://civarchive.com/models/120475/cornelia-li-britannia-locon-version-code-geass">the LOCON version </a>which is why that came out first. I normally put LOCONs first on the queue as they are more sensitive and if i screwed up they show it very easily. Regardless Let's get to the important bits.V2:Works fine with: &lt;lora:cornelia_li_britanniaV2:0.7&gt; cornelia_li_britannia,The outfits I managed to isolate are:<ul><li>cornelia_military_uniform: Her classic attire.</li><li>cornelia_white_cape: Her weird cape that goes with her classic attire.</li><li>white_bodysuit_red_bustier_red_gloves_red_thigh_boots: her secondary casual attire. I lightly trained her other cape but YMMV is is just tagged as cape.</li></ul>As Always I will upload the tag summary with the dataset for people to experiment.V1:Another quality waifu. I am trying to focus on side characters as the most popular ones will inevitably be picked by someone else.I got good results with: &lt;lora:Cornelia_li_Britannia:0.6&gt; Cornelia_li_Britannia, 1girl, big_breasts, purple_eyes, (lipstick:.9), (makeup:.8)It seems in this case a lower CFG turned out better and ended up using 6.Note: I think she has mole in roughly 25% of the training data so you might get it if you add the tag.

Files with SHA256: 1e3725205262f6037781137d8fc92d47966067d2e06f0ff11cb2a4ba83e8ff48