This is a very early prototype and has faults which I will explain below.This model has been trained similar to my <a target="_blank" rel="ugc" href="https://civarchive.com/models/704013/photorealistic-nsfw">other model</a>, but with added xxx images including blowjob, penetration (anal and pussy), bondage, cumshots and more.The best way to use this model is to use the comfyui nodes that are attached to the sample images.The workflow is a 3 step process:<ul><li>Generate an image with Flux and this LoRA.</li><li>Do img2img with an sdxl model. I use <a target="_blank" rel="ugc" href="https://civarchive.com/models/575395/big-lust">bigLust</a>.</li><li>Enhance face, eyes and lips. This image is not always the best and sometimes image from the step above is the best.</li></ul><h3 id="training-jr51fj9mh">Training</h3>I used aitoolkit.lr: 0.0001steps: 50Klr_scheduler: cosinerank: 64/64<h3 id="captions-ichcc6h0e">Captions</h3>I took about 250 images of nude women in different poses, lighting that were captioned with a mix of joycaption and wd-vit-large-tagger-v3, manually shuffled and inspected to remove unneeded words or sentences.I then added another 250 xxx diverse images featuring blowjobs, bondage, penetration, cumshots, etc that were only tagged with wd-vit-large-tagger-v3.A total over 500 images were used.I found that only tagging the xxx images without descriptive text does not give the results I wanted and I'll probably need to re-run through the captioning for these like I did for the other images and re-run training.For now, use tags and descriptive captions to get the best results.

Files with SHA256: f0dcb9dc729c460e50d15cb480117c7dacff9389f2b327a6dc84d51d89bb357d