<h2 id="heading-2">Posted for <a target="_blank" rel="ugc" href="http://civarchive.com">civitai.com</a> and <a target="_blank" rel="ugc" href="http://tensor.art">tensor.art</a> only</h2>Each of these LoRAs take me dozens of hours and GPU usage out of pocket. Every review is appreciated and encourages me to do more of these.-MoCSee about this version for model details 👉<h1 id="heading-28">💃Mila Azul v2 Upgrade💃</h1><h2 id="heading-29">v2 Changelog</h2>I've started from scratch for this v2 LORA.<ul><li>Newer higher quality dataset, with only the choicest and ideal images.<ul><li>More consistency</li><li>More pose flexibility</li></ul></li><li>Trained at much higher resolution to take advantage of the increased quality.<ul><li>More details</li><li>Better more consistent upscales/inpaints</li></ul></li><li>Using my updated training methodology.<ul><li>No random settings or pray and spray approach for this one.</li></ul></li><li>To give an idea of the increase in quality and resolution: This took nearly 2hrs to bake, versus my usual sub 30 minute cook!! 👀</li></ul><h1 id="heading-30">‼ IMPORTANT ‼</h1><ul><li>Your voice matters! Show your support and leave a post, comment and most importantly a review! ⭐</li><li>Custom LORAs available upon request. Just buymeacoffee.☕</li><li>(workflow advice below) 👇</li></ul><h1 id="heading-31">📝Biography📝</h1><h2 id="heading-32">🌟Mila Azul🌟</h2>whose birth name is Ekaterina Volkova, is a Ukrainian model and musician born on December 1, 1997 in Kiev. She stands at a height of 5'7" (1.70 m) and is known for her buxom and shapely figure, with large natural bosom and a slender build. Mila began her modeling career in 2016 at the age of nineteen and has since worked with various high calibre modeling websites. In addition to modeling, Mila is also a musician. She has stated that she works in Europe, lives in Ukraine, and is open to collaborating with new people.<h2 id="heading-33">💡Workflow Advice💡</h2><h3 id="heading-34">🔧Prompt and Sampling🔧</h3><ul><li>Most images have been generated with LORA weighted @ .75 - 1.</li><li>For photoreal shots, I've found sampling steps best well above 50 give better results.</li><li>ControlNet is your friend. Embrace the dark side.😈<ul><li>Tile pre-sample and model is good for extreme close ups</li><li>I often use segmentation for my generations now</li></ul></li></ul><h3 id="heading-35">🚀Upscaling🚀</h3><ul><li>Try the <a target="_blank" rel="ugc" href="https://github.com/Coyote-A/ultimate-upscale-for-automatic1111">Ultimate SD upscale script</a>!</li><li>Sometimes I just send it to Ultimate SD upscale after a quick render and run 2x at a scale of two.</li><li><a target="_blank" rel="ugc" href="https://github.com/Coyote-A/ultimate-upscale-for-automatic1111">Ultimate SD upscale</a> script used @ ~.15 - .4 denoising using 8x NMKD Faces or 4X UltraSharp redraw upscalers.</li></ul><h1 id="heading-36">🗒️Side Notes🗒️</h1><ul><li>Help me help you: Leave a review if you've used this LORA.</li><li>Most of my gens are not voodoo or outside of workflows I've indicated here or in my other posts.<ul><li>In other words: Leave a comment if you are having issues getting good output, I'll try to respond with some suggestions.</li></ul></li></ul>

Mila Azul + XL

<ul><li>Dataset gathered and trained on a single concept, only from 2016.</li><li>~8 hour Kohya_ss training in 16200 steps across 27 epochs @base dim (rank): 32.</li><li>1248x1248px training resolution base</li><li>Common bucketed resolutions @ 1024x1472, 1216x1216</li></ul>

Files with SHA256: 2d8e6146b4c5e1646a358407f5c32c5ec1b19826691c322683dc6518e536bf81