Trigger Word: <code>b3@ut1f0ll_r@in</code> Model: Wan 2.1 t2i 1.3B All examples are generated with 1.0 LoRA strength and CFG=6 For inference used <a target="_blank" rel="ugc" href="https://github.com/kijai/ComfyUI-WanVideoWrapper/tree/main">Kijai's workflows</a><h3 id="training-details-ij0j1pkoc">Training Details</h3>This is my first attempt to teach lora for Wan 1.3B such a specific movement as painted rain. The dataset included 11 videos from the anime Kotonoha no Niwa (The Garden of Words) with lengths of 16, 30, 60, 72 and 81 frames. <a target="_blank" rel="ugc" href="https://github.com/tdrussell/diffusion-pipe">Diffusion-pipe</a> was used for training. The training was conducted simultaneously at resolutions 256x456 and 480x854 with num_repeats=1. I used a slightly higher learning rate and as the best model I chose the 54 epoch with lr=8e-05 with LoRA rank of 32.The version for Wan 14B can be found here <a target="_blank" rel="ugc" href="https://civarchive.com/models/1747192/beauty-of-rain-wan-14b">https://civarchive.com/models/1747192/beauty-of-rain-wan-14b</a>

Files with SHA256: 6b45b643c9bb42264f6bf7a955c74c8721bf5f8d0263b8752bd9c284ba3cc142