Workflow embedded in images.Trigger words are:<ul><li>selfplay. licking her nipples.<ul><li>for the licking photos</li></ul></li><li>selfplay. sucking her nipples<ul><li>for the sucking photos</li></ul></li></ul>For training I used the settings added in the zip in the resources section for <a target="_blank" rel="ugc" href="https://github.com/tdrussell/diffusion-pipe">https://github.com/tdrussell/diffusion-pipe</a>I trained it on a 5090 and took a couple of hours if I remember correctly.The dataset was 80 images (40 licking / 40 sucking) captioned with the ollama captioner workflow from here: <a target="_blank" rel="ugc" href="https://civarchive.com/articles/22754">https://civarchive.com/articles/22754</a>I found that this training works really well for training concepts on Chroma1 HD -- not good for character lora training though so don't try it. For good character lora training you need more repeats / image (around 8 with a learn rate of 2e-4).

Files with SHA256: 087cfa97d5d0f5add08d3aff233290bf920ceb15eea041b17841069f0a86b56f