Makes images similar in style to the artist <a target="_blank" rel="ugc" href="https://twitter.com/BBCChanArt/">BBC-Chan</a>. LyCORIS version of an old <a rel="ugc" href="https://civarchive.com/models/100504/bbc-chan-style">LoRA</a> with a slightly refined dataset and trained with more up-to-date methods. Unlike the previous release this one is trained on NAI which should result in better compatibility with any model that's based on it. How to use Just prompt away, no trigger word needed. Refining the dataset ended up with one that leans mainly towards tags like "dark-skinned female", "pointy ears", "muscular female", various animal and monster girl tags as well as feminine penises. It also tends to prefer black and tight clothing unless prompted otherwise. While I haven't tested it much beefcake males should work fine too. Recommended strength is anything from 0.5 to 1.0 depending on what you prompt. As grid image shows you can even prompt just "1girl" and get something reasonably in style. Models with some realism mixed into them like AOM2 generally works best with height or width at 768px. Lower values work too but faces tend to end up worse. Alternatively, using lower resolution and then upscaling or inpainting in img2img works too. For more prompt tips see the sample image or the uploaded zip with training tags.MISC Trained on a dataset of 59 images at 512 resolution on NAI. All manually cleaned and upscaled. All sample images are raw txt2img upscaled, no helper embeddings/negatives used. E.g. - things should look even better if you do. The training was done on a small subset of images, as such the output of this LyCORIS is not an accurate representation neither in quality or style of the source material. Go support the artist if you want the good stuff.

Files with SHA256: 2a06067d9e082003c0c4743bbb9c53c6ef9b8267d60a2c1a0c91914fa852c26f