<h1 id="heading-4505">[LoKr + LoRA] Venus Park</h1><h3 id="heading-60">Notice:</h3><p>This model is more for experiments and is not well trained. (it may need extra tags/prompting to reproduce some characterstics)</p><p>You should refer to all the example images to check if "the lacking of some features" is due to lacking of some tags.</p><p></p><h3 id="heading-265">Usage:</h3><p>Put this model in to models/lora folder of sd-webui. And use kohaku-xl beta series model to use it.<br />(This model is trained on sdxl1.0 and I recommend user to use kohaku-xl beta series model)</p><p></p><h2 id="heading-61">Introduction</h2><h3 id="heading-62">An experiment model for:</h3><ul><li><p>LyCORIS preset config system(using different algo/setting for different modules)</p></li><li><p>LyCORIS character model on SDXL arch</p></li><li><p>FP8 training for LyCORIS</p><ul><li><p>I trained this model with only 8GB vram. (Without cache TE)</p></li></ul></li><li><p>FP8 inference on sd-webui</p><ul><li><p>I implemented some FP8 opt in sd-webui <code>test-fp8</code> branch and all the example images are generated with FP8 UNet/TE</p></li><li><p>Only take 4GB vram when generating the images I post in this model</p></li></ul></li></ul><h3 id="heading-63"><br />Training settings:</h3><ul><li><p>Batch size: 1</p></li><li><p>Epoch: 50(without any repeat)</p></li><li><p>LR: 2e-4/5e-5</p></li><li><p>Optimizer: AdamW8bit(weight_decay=0.02, betas=(0.9, 0.95))</p></li><li><p>Bucket Reso: 1024x1024(448~2112)</p></li><li><p>shuffle caption/caption dropout enabled</p></li></ul><p></p><h3 id="heading-64">More infos</h3><p>For FP8, I will publish a blog(or note?) for FP8 training and inference recently. Stay tuned.</p><p>For preset config system, plz refer to the docs of <a target="_blank" rel="ugc" href="https://github.com/KohakuBlueleaf/LyCORIS">LyCORIS</a></p><p>For more informations: just comment below.</p><p><br /><br /></p>

Files with SHA256: bc276d1770c52498cc88757671692a9c2685670e07e83ad955a3fcfa4b0933a9