<h3 id="cowgirl-concept-lora-(video-or-wan2.2-5b)">🤠 Cowgirl Concept LoRA (Video | WAN2.2 5B)</h3>Hey everyone! This is my first video LoRA, trained for cowgirl-style grinding/riding animations using [WAN2.2 5B]. I followed the official <a target="_new" rel="ugc" href="https://www.youtube.com/watch?v=9ATaQdin1sA">YouTube tutorial</a> and used <a target="_new" rel="ugc" href="https://github.com/ostris/ai-toolkit">Ostris’ AI Toolkit</a> — surprisingly easy and works on windows.I used the same settings shown in the video.<hr /><h3 id="training-details:">🔧 Training Details:</h3><ul><li>Model Target: WAN2.2 5B</li><li>Dataset: AI-generated 3-second videos (72 frames each) made with WAN2.1</li><li>Includes hair physics and breast motion, though the dataset wasn’t perfect</li><li>Training Steps: 3000 total, but best results around step 2500</li><li>Training Time: ~7 hours on an RTX 5090</li><li>VRAM Usage: peaked around 16GB</li><li>Workflow: Used the default ComfyUI workflow, just added a LoRA loader — worked flawlessly</li><li>Trigger Word: <code>c0wg1rl</code></li></ul><hr /><h3 id="notes:">💡 Notes:</h3><ul><li>Could likely train faster with proper batch settings (I didn’t use one this time).</li><li>While WAN2.2 5B doesn’t match the quality of the 14B model, it’s much faster and more resource-efficient, and still delivers good results.</li><li>Definitely room for fine-tuning and dataset refinement in the future.</li></ul><hr />Let me know what you think, and feel free to share any tips! Hope this helps others get started too 🎬🐮

Files with SHA256: c1fbbdbf5aa4a3a88b82804365a8114fa0556c3d0c3e41a20c0b2c5b855de09c