<h2 id="v2-is-here!-4ocsqv5nh">V2 is here!</h2>Differences:<ul><li>It is a bit weaker,</li><li>follows the prompt better,</li><li>Less bleeding to the other characters</li></ul><h2 id="inference-0yiqpq3ma">Inference</h2>It seems, the model likes long descriptive prompts. Look at the attached videos for prompt examples. Based on my tests if you use short prompts, the lora effect is weak.For inference I recommend:<ul><li><a target="_blank" rel="ugc" href="https://blog.comfy.org/p/wan21-video-model-native-support">ComfyUI native nodes</a></li><li><a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-WanVideoWrapper">Kijai's ComfyUI wrapper nodes</a></li></ul>Trigger words: csetiarcane, Nfj1nx, blue hair(If it isn't enough so you don't get the style/character, I recommend adding "animation style" to you prompt. It can help providing the style but in some cases the result will be too cartoony)Strength: 0.9-1.0<h2 id="trainig-details-(v1):-ntnt6gxj5">Trainig details (v1):</h2>I tried a method where I first trained the LoRA on images only and then after a certain number of steps continued training only on videos.<ul><li>LR: 2e-5</li><li>Optimizer: adamw_optimi</li><li>epochs: 15 (image) + 7 (video)</li><li>steps: 2550 (image) + 2520 (video)</li><li>dataset: 33 images and 35 videos</li><li>repeats: 10</li><li>rank: 128</li><li>batch size: 2</li><li>gradient accumulation steps: 1</li></ul>For training I used the <a target="_blank" rel="ugc" href="https://github.com/tdrussell/diffusion-pipe/tree/main">Diffusion-Pipe</a> repo.<h2 id="acknowledgement-9ijf30w7m">Acknowledgement</h2>Special thanks to <a target="_blank" rel="ugc" href="https://github.com/kijai">Kijai</a> for the tireless and excelent work done for the community to enable us to use these solutions as soon as possible, as well, as to <a target="_blank" rel="ugc" href="https://github.com/comfyanonymous">Comfyanonymous</a> for implementing them so quickly with such a good quality natively into ComfyUI, and also to <a target="_blank" rel="ugc" href="https://github.com/tdrussell">TDRussel</a> for making LoRA training available to the community so quickly.

Files with SHA256: a48d5d95361c2d3c97bfbdad297c89879378a9b0a4c508c6871ae300b2f9fed4