New for v2!<ul><li>Trained on SD1.5 base</li><li>Larger dataset.</li><li>All-new dataset (generated from a different prompt). Full set at <a target="_blank" rel="ugc" href="https://huggingface.co/datasets/ppbrown/faeryqueen">https://huggingface.co/datasets/ppbrown/faeryqueen</a></li></ul><h2 id="heading-158">Important notes</h2><ul><li>For some odd training reason, putting "red eyes" in the negative, often normalizes things. Your choice of model determines whether it is needed or not.</li><li>Some models do best when you drop to cfg=4. Others do not.</li></ul>This is a fairly simple LoRA. Sort of like an embedding, really. Pushes a character towards the style of a "Faery Queen". Elegant ethereal female with wings.Not a particularly amazing LoRA. I made it primarily to learn how to make them. If you would like to know how I made it, check out my article!<a target="_blank" rel="ugc" href="https://civarchive.com/articles/4789">https://civarchive.com/articles/4789</a>Interestingly, I split it into two "concepts", and now it requires a trigger word "faery", whereas before, it didnt really need it<h2 id="heading-158"></h2>

FaeryQueen SD

I tried doing a buncha different things this time. one of which is training on SD1.5 BASE model! Plus, drastically changed my prompt for source images. Lemme know what you think.

Files with SHA256: 6cf7c119aef6feb17cc96c00e048d2a563954afdc0a647413b23259d47229367