<p>Generates the character in Jon-Yakitory's <a target="_blank" rel="ugc" href="https://www.youtube.com/watch?v=1Swg-aBO9eY">Konton Boogie</a>.</p><p>Also generates images closely resembling Jon-Yakitory's style, albeit you have to ditch "Konton"'s specific tags.</p><p></p><h2 id="tested-on:">Tested on:</h2><p><a target="_blank" rel="ugc" href="https://civarchive.com/models/2754368/anima-int8-mxfp8?modelVersionId=3099040">Anima Base v1.0 int8</a></p><h2 id="showcase-images-settings:">Showcase images settings:</h2><p>Sampler: LCM (haven't tested the other samplers, but they should also work fine)</p><p>Sched: sgm_uniform</p><p>CFG: 1.0 (again, works fine &gt;1.0)</p><p>Resolution: 512x768 -- Upscaled 4x with <a target="_blank" rel="ugc" href="https://openmodeldb.info/models/2x-StarSample-V2">StarSample V2</a>.</p><h2 id="important">Important</h2><p>You must use the <a target="_blank" rel="ugc" href="https://civarchive.com/models/2560840/anima-turbo-lora">Anima Turbo LoRA</a> if you want to replicate the style precisely, otherwise it'll just look buns.</p><p></p><hr /><h2 id="training-info">Training info</h2><p>LR: 1e-4</p><p>Network Dim/Alpha: 16/16</p><p>Optimizer: AdamW8bit</p><p>Steps: 400</p><p>Images: 11</p><p>Size: ~480²</p><h2 id="time-taken">Time taken</h2><p>Dataset preparation + tagging: 3min</p><p>Training: 24min</p><p>Total: 27min</p>

Files with SHA256: f51f5b1f966227a2893fbe38fdb7478f4335ea03f76fccc85be9fcba9f36ba65