Dolly Effect

🚀 万2.1 14B I2V 480p的移动变焦效果LoRA

这个LoRA模型基于万2.1 14B I2V 480p模型训练，能为任意图像主体添加移动变焦（Dolly zoom）相机效果！该模型也适用于文本到视频（T2V），提示词风格与图像到视频（I2V）类似，但I2V的应用效果更出色。

🚀 快速开始

本项目提供了一个可用于图像到视频转换的LoRA模型，能为视频添加移动变焦效果。你可以按照以下步骤开始使用：

1. 下载模型文件和推理工作流文件。
2. 按照推荐设置进行参数调整。
3. 使用触发词和提示词模板生成视频。

✨ 主要特性

• 基于特定模型训练：基于万2.1 14B 480p I2V基础模型训练。
• 结果一致性高：在不同对象类型上都能获得一致的效果。
• 提示结构简单：提示词结构简单，易于调整。

📦 安装指南

📥 下载链接

• dolly_25_epochs.safetensors - LoRA模型文件
• wan_img2vid_lora_workflow.json - 适用于ComfyUI的万I2V带LoRA工作流文件

💻 使用示例

示例提示词及输出

• 提示词：d011Ye33ect dolly effect. The video begins with a close-up of the man’s steely gaze as he stands in a dusty cemetery, a cigar clenched in his mouth. The camera slowly zooms out, keeping his face centered while the background stretches—revealing crosses, gravestones, and the wide open desert behind him. The dolly effect intensifies the tension of the western standoff. 输出视频：example_videos/1.mp4
• 提示词：d011Ye33ect dolly effect. The video begins with a close-up of the woman's face, her expression calm and confident. As the camera zooms out slowly, her poised figure in the iconic seated position is revealed. The background — a sterile, cold interrogation room — subtly distorts with the dolly effect, enhancing the intensity and drawing focus to her unwavering gaze. 输出视频：example_videos/2.mp4
• 提示词：d011Ye33ect dolly effect. The video begins with a close-up of the man’s intense expression, his mouth open mid-shout. As the camera slowly zooms out, his battle stance and outstretched arms are revealed in full. The dolly effect causes the background of the ancient coliseum to shift and distort slightly, emphasizing the tension and power of the moment. 输出视频：example_videos/3.mp4
• 提示词：d011Ye33ect dolly effect. The video starts with a close-up of the dogs’ faces as they share a single strand of spaghetti. The camera slowly zooms out, revealing the candlelit table, checkered tablecloth, and surrounding alleyway. The dolly effect keeps the dogs centered as the background stretches subtly, enhancing the intimacy of the moment. 输出视频：example_videos/4.mp4

推荐设置

触发词

关键触发短语为：d011Ye33ect dolly effect

ComfyUI工作流

此LoRA模型可与 Kijai的万视频包装器工作流 的修改版本配合使用。主要修改是添加了一个连接到基础模型的万LoRA节点。 修改后的工作流文件可在上述下载部分获取。

📚 详细文档

模型信息

模型权重以Safetensors格式提供，具体下载链接见上述下载部分。

训练详情

额外信息

训练使用了 Diffusion Pipe for Training。

致谢

特别感谢Kijai提供的ComfyUI万视频包装器，以及tdrussell提供的训练脚本！

社区

• 加入社区：加入我们的社区，免费使用此LoRA模型生成视频。
• 请求LoRA模型：我们正在免费训练并开源万2.1 LoRA模型 - 加入我们的Discord提出请求！

📄 许可证

本项目采用Apache-2.0许可证。