Greetings!Following the recurring theme of not finding exactly what I'm looking for, I've decided to do one myself. This model is still on alpha and currently doesn't work well on realistic images and on darker skin tones. This is mostly because of the small dataset that I'm working with (~60 images). This model will most likely get a full release in a few weeks fixing those problems and improving further on quality.<h1 id="heading-1708">Model Characteristics</h1><ul><li>Trained on 512x512 but can produce good results on 512x768 too.</li><li>Trigger word: queen of hearts symbol</li><li>Use weight of 0.8 to 1. See what works best for you.</li><li>I'm using <a target="_blank" rel="ugc" href="https://civarchive.com/models/6174">DivineEleganceMix </a>and <a rel="ugc" href="https://civarchive.com/models/19831?modelVersionId=30151">DegenerateHassakuV1</a> and got good results, so I'd recommend using checkpoints that have an anime style.</li><li>Play around with the samplers, but I got best results with Euler a.</li><li>Clip Skip 2 works best.</li><li>Hires Fix is strongly recommended since it enhances the shape of the tattoo.</li></ul><h1 id="heading-1709">Trainning</h1>I'm relatively new to LoRa trainning so any tips for improvement are much appreciated!<ul><li>Dataset size: 57</li><li>Model: <a target="_blank" rel="ugc" href="https://civarchive.com/models/23900">AnyLora</a></li><li>Image size: 512x512</li><li>Epoch: 1</li><li>Batch size: 2</li><li>Repeats: 100</li><li>Total steps: 2850</li><li>Trainning time on GTX 1070 8GB: ~3 hours</li></ul>Like my stuff? Buy me a Monster!<a target="_blank" rel="ugc" href="https://ko-fi.com/lizarb"><img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/1ae16ea7-0a66-4765-a795-b9681a98a4ae/width=525/1ae16ea7-0a66-4765-a795-b9681a98a4ae.jpeg" /></a>All my models and images will always be 100% free, don't feel obligated to donate!

Files with SHA256: 15c51f6db070264f7cd6abd8f497ff0adfb23f5ff45c1cb5099e717e9477216b