After some experimentation with merging different models together, I've got a decent result on a few. The idea for these checkpoint merges was to work good with the Loras I've trained and was going to train.The merge should output good results for both characters and landscapes. I see the style as a semi-realistic, but leaning more towards the anime side. Experiment.Hires. fix is recommended (the higher the better, 2 times is best, but <a rel="ugc" href="https://i.imgflip.com/igpav.jpg">VRAM</a> is <a rel="ugc" href="https://pbs.twimg.com/media/FzxAE37aYAAQiwZ?format=jpg&amp;name=900x900">hard to come by</a>, I feel your pain...)If you (more like your GPU) are struggling with upscaling with Hires. fix, as an alternative I can recommend <a rel="ugc" href="https://github.com/Coyote-A/ultimate-upscale-for-automatic1111">Ultimate SD upscale</a>.<a rel="ugc" href="https://github.com/Bing-su/adetailer">ADetailer</a>. It should be used if you want a character's face to be detailed and not funky-looking. It could also be used to add a certain face to a character (by placing the Lora of said face in the ADetailer's prompt).What I use as parameters for generating:<ol><li>DPM++ 2M Karras (Euler a and SDM work too, but 2M is the most consistent)</li><li><a rel="ugc" href="https://civarchive.com/models/116225/4x-ultrasharp">4x-UltraSharp</a> as the Upscaler (others work fine, too, use whichever you prefer)</li><li>Sampling Steps are anywhere between 25 and 30</li><li>Hires steps are 15</li><li>CFG anywhere from 7 to 10</li></ol>

Files with SHA256: 63d2dc9c4259d3d19729f005a68e22dfce9cfff76b518451b4959571a96858ef