<h2 id="my-first-wan-2.2-lora">My First Wan 2.2 LoRA</h2>This is my first attempt at training a video LoRA.It is not perfect, but it works—and for me, that already makes it a meaningful first step. The LoRA was trained on Wan 2.2, which I still find to be a reliable foundation for character-focused video generation.I also experimented with LTX 2.3, but in my experience, it still struggles with character consistency and identity preservation. Facial features can become unstable or distorted, especially during larger movements or longer generations. Because of this, I decided to continue training with Wan 2.2 for now.Hopefully, more powerful and consistent open video models will arrive in the future. I am looking forward to seeing how the ecosystem develops—and to training better versions as I gain more experience.<h3 id="trigger-words">Trigger Words</h3>No trigger word is required.<h3 id="prompting">Prompting</h3>The LoRA is designed around scenes featuring two people.And the prompting just : kiss .Example prompt:<code>two people kiss.</code><code>They are kissing.</code>For better results, try adding:<ul><li>Detailed descriptions of both characters</li><li>Clothing, hairstyle, and facial features</li><li>Camera movement and shot composition</li><li>Lighting and environment</li><li>The specific interaction or action</li><li>Mood, pacing, and visual style</li></ul>Since this is my first LoRA, results may vary depending on the base model, resolution, sampler, LoRA strength, and generation workflow.Thank you for trying it. Feedback, example videos, and workflow suggestions are always welcome.

Files with SHA256: b884cbb0c129a2bb11f48e59ada4cc4489592da74d215aef09413edfed4c5309