AnimateDiff 运动生成测验

当您参加测验时，就会出现答案和解释。

1. AnimateDiff 背后的核心思想是什么？
   • 在数百万个剪辑上从头开始训练全新的视频模型
   • 仅使用光流在两个图像之间进行插值
   • 录制真实视频并逐帧重新设计
   • 将经过训练的运动模块插入冻结图像扩散模型
2. 重用 AnimateDiff 运动模块时，哪些检查点兼容性条件很重要？
   • Every image generator accepts every motion module regardless of architecture
   • The personalized image checkpoint must be compatible with the supported base architecture and module
   • Freezing any arbitrary model makes it automatically compatible
   • The image checkpoint's filename alone establishes compatibility
3. 运动模块如何保持帧之间的一致性？
   • 通过时间注意力层，让帧跨时间参与
   • 通过均匀地模糊每一帧
   • 通过仅生成一帧并复制它
   • 通过使用固定的预先录制的动画
4. AnimateDiff 与扩展最相关的是哪个模型系列？
   • GPT 语言模型
   • 语音识别模型
   • 稳定的扩散图像模型
   • 表格回归模型
5. AnimateDiff 生态系统中的运动 LoRA 通常控制什么？
   • 文本编码器词汇
   • 特定的摄像机移动，例如平移、缩放或滚动
   • 视频的音轨
   • 输出的文件格式
6. How should a sixteen-frame AnimateDiff result be interpreted relative to the method's possible configurations?
   • As a guarantee that every deployment must output exactly sixteen frames
   • As proof that the method generates a feature-length film in one step
   • As a rule that the generated video contains only a single still image
   • As a short-clip configuration, rather than a universal fixed duration for every version
7. What does SparseCtrl add to an AnimateDiff-style generation workflow?
   • A requirement to provide a manually labeled image for every frame
   • An audio transcription model that replaces temporal attention
   • Conditioning from a sparse set of reference images at selected frame positions
   • A guarantee that any unrelated image architecture is compatible
8. How does a motion LoRA's camera control differ from SparseCtrl's reference-image guidance?
   • A motion LoRA steers a learned movement pattern, while SparseCtrl supplies visual content at selected frames
   • They are identical names for a text tokenizer
   • SparseCtrl controls only the video filename
   • Motion LoRAs require a complete reference frame at every timestamp