I created a Lora based on a dataset of 72 images of black and white hearts in Photoshop, just to see what happens. Got some interesting results. Trigger word is: BWHeart. Also here is the list of tags that appeared in the captions, so using these words in the prompt or negative prompt could also have a substantial effect on the outcome. <img src="https://imagecache.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/3e2b65c3-f2d3-4332-d5de-95e2cb3c4200/width=525/3e2b65c3-f2d3-4332-d5de-95e2cb3c4200.jpeg" /> Models react to the Lora quite differently, and some require less power to make anime characters, while some require more to bring out the Lora's abstractness. So hard to give instructions. It also has a tendency to create multiple girls, even with "multiple girls" and "multiple views" put into the negative prompt. But this makes it pretty good at making yuri images too! If you like this one, you might also like my other Lora that is similar to this here: <a target="_blank" rel="ugc" href="https://civarchive.com/models/41046/bwgeoabstract">https://civarchive.com/models/41046/bwgeoabstract</a>

Files with SHA256: b9e9f0b899c45d288c6aec6cf19681b67649eb87e57756e9d36eedcea60d770e