Training based on <pre><code>wan2.1_t2v_14B_fp16.safetensors</code></pre>This model is used to restore the character nahida in the Genshin Impact,This is the third version of the test, which has good details and natural movements,But it did not achieve the best effect I expected and is still being debuggedReference for prompt words:<pre><code>A digital drawing in an anime style features Nahida, She has silver hair that transitions into green gradient tips, adorned with cross-shaped patterns in her pupils. Her long multicolored hair includes a complex cross-shaped metal edging on one side and braided hair with green highlights on the other, ending in a bright green leaf-shaped accessory.Nahida wears a unique sleeveless dress primarily white, featuring an inverted heart at the chest with a central glowing emerald hourglass totem. The bottom of the dress has intricate flower bud-shaped patterns, and the hem is adorned with green droplet-shaped decorations and metal edging. She complements this outfit with jewelry including a bracelet. The texture is smooth, featuring clean lines and vibrant colors typical of high-quality digital art.</code></pre><pre><code>Nahida ,In this whimsical video, we follow Nastia the fairy as she playfully explores a vibrant meadow filled with wildflowers and dappled sunlight. The camera pans gently from her tiny figure to capture her mischievous twirls among the flowers, each petal appearing to dance around her in a shimmering glow. Nastia uses her delicate wand to summon playful sparkles that float around her like tiny stars, creating a magical whirlwind of light. As she giggles and hops from one flower to another, the camera focuses on her sparkling eyes, which shine with innocence and wonder. The scene is enhanced by soft fairy dust trails left in her wake, adding an enchanting touch as they linger briefly before fading into the warm breeze.</code></pre>The prompt words are divided into two parts, the first part is responsible for describing the characteristics of the character, and the second part is responsible for describing the action situation,The former is a description used during training, while the latter is a rich action prompt generated by LLM model.Test resolution: 640x640steps：20cfg：8shift：5scheduler：dpm++—————————————中文说明，内容同上—————————————训练基于：<pre><code>wan2.1_t2v_14B_fp16.safetensors</code></pre>这个模型用于还原原神里的角色 纳西妲，这是测试的第三个版本，有着较好的细节和较为自然的动作，但并未达到我期望的最佳效果，还在调试中对AI训练感兴趣的欢迎加入我的AI交流QQ群：839674395交流训练技巧~提示词参考<pre><code>A digital drawing in an anime style features Nahida, She has silver hair that transitions into green gradient tips, adorned with cross-shaped patterns in her pupils. Her long multicolored hair includes a complex cross-shaped metal edging on one side and braided hair with green highlights on the other, ending in a bright green leaf-shaped accessory.Nahida wears a unique sleeveless dress primarily white, featuring an inverted heart at the chest with a central glowing emerald hourglass totem. The bottom of the dress has intricate flower bud-shaped patterns, and the hem is adorned with green droplet-shaped decorations and metal edging. She complements this outfit with jewelry including a bracelet. The texture is smooth, featuring clean lines and vibrant colors typical of high-quality digital art.</code></pre><pre><code>Nahida ,In this whimsical video, we follow Nastia the fairy as she playfully explores a vibrant meadow filled with wildflowers and dappled sunlight. The camera pans gently from her tiny figure to capture her mischievous twirls among the flowers, each petal appearing to dance around her in a shimmering glow. Nastia uses her delicate wand to summon playful sparkles that float around her like tiny stars, creating a magical whirlwind of light. As she giggles and hops from one flower to another, the camera focuses on her sparkling eyes, which shine with innocence and wonder. The scene is enhanced by soft fairy dust trails left in her wake, adding an enchanting touch as they linger briefly before fading into the warm breeze.</code></pre>提示词分为两个部分，第一部分负责描述角色特征，第二部分负责描述动作情况，前者为训练时采用的描述，后者是用AI大语言模型生成的丰富动作提示。测试分辨率：640x640steps：20cfg：8shift：5scheduler：dpm++

Nahida_Wan2_1_Lora

优化了提示词和训练集，增加了泛化

Files with SHA256: 27c686dd20ac986991eaebe165684b724d01cd1d3646889aead7147b89426f60