LoRA for Depresso, a dark element Pal from Palworld.Recommended weight: 1.0Main trigger word: depressoFor "default" look: blue body, purple hair, pink eyes, black tailFeel free to adjust colors to get your own unique Depresso:<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/5bab9db2-c99d-4d3d-ba70-4798e11297e0/width=525/5bab9db2-c99d-4d3d-ba70-4798e11297e0.jpeg" /><h3 id="version-2.1-tbvfcgj57">Version 2.1</h3>I swear, this is the last one.Did minor tweaks to the dataset to fix issues that popped up in version 2.Also changed network_dim to 16 from 32 to lower LoRA size.Overall I'm very satisfied with results and I can say that it's a definitive improvement over the version 1.<h3 id="version-2-1lo053iro">Version 2</h3>Decided to redo this LoRA by reworking captions and tweaking dataset a bit. Not a major improvement (if there is an improvement), but seems to be more flexible, at least in some ways.Known issues:<ul><li>messy eyes</li><li>"collar" sometimes doesn't appear fully black</li><li>bizarre forest</li></ul><h3 id="version-1-p7zj55lwp">Version 1</h3>Initial releaseMy first serious attempt to use LoRA training on this website. The model turned out to be surprisingly decent, despite the fact that training data was kinda a mess.

Depresso (Palworld) | PonyXL

<ul><li>10 epochs, 480 steps;</li><li>Recommended weight 1.0;</li><li>Trigger word: Depresso;</li><li>blue body, purple hair, pink eyes, black tail for "default" look, colors can be changed (usually works well).</li></ul>Minor tweaks in dataset in attempt to fix issues that popped up in v2.0Also changed dim to 16, as default 32 seemed to be too extreme for a character like this.

Files with SHA256: 0ab747391ef7e0660040c923bb93247dd2414ef0a558ac6e016b2d40d16e8a39