<p>PonyXL style LoRA for emulating laserflip's art style. Works well with base model and AutismMixSDXL.</p><h3 id="update-31024:-v1.1-release-cfgnn6d45"><span style="color:rgb(250, 176, 5)">Update 3/10/24: V1.1 release</span></h3><p>New version trained on updated dataset.</p><p></p><hr /><p>Baked for 20 epochs for a total of ~2000 steps on a thoroughly cleaned dataset - no text or watermarks here.</p><p></p><h3 id="usage-qe374qre6">Usage</h3><pre><code>Activation Tag: laserflip
Weight: 0.7~1.0</code></pre><h3 id="notes-imtvnfmn2">Notes</h3><ul><li><p>This model has a heavy NSFW bias (duh). SFW is doable with <code>rating_safe</code> tag.</p></li></ul><ul><li><p>The dataset contained a not insignificant portion of censored images. Include <code>uncensored</code> in positive prompt or <code>mosaic censoring</code> in the negative prompt if your outputs are censored.</p></li></ul>

[PonyXL LoRA/Style] Laserflip Style

<ul><li><p>Retrained on revised dataset (old dataset had glaring tag issues)</p><ul><li><p>Improved pose variety</p></li><li><p>Fixed overly simple oil paint looking backgrounds</p></li></ul></li><li><p>Increased network dim 16 --&gt; 32 (should retain details better)</p></li></ul>

Files with SHA256: 784ee67aacd013b3834d3f8d7874f4914246c8af0de17a87850df415dc722b86