Imagen Video Cascades Quiz

Answers and explanations appear as you take the quiz.

1. Which design helps Imagen Video represent both spatial detail and changes across frames?
   • A Video U-Net that uses temporal self-attention in the base model
   • A frozen spreadsheet encoder with no video operations
   • A speech-only recurrent network
   • A single still-image classifier applied once
2. What does a temporal super-resolution stage do in the cascade?
   • Increases the pixel resolution
   • Adds more frames between existing ones
   • Translates the prompt
   • Removes the audio track
3. What component encodes the text prompt in Imagen Video?
   • A convolutional image classifier
   • A speech recognition model
   • A frozen T5 language model
   • A reinforcement learning agent
4. Why did Imagen Video use noise conditioning augmentation in its super-resolution cascade?
   • Increase pixel resolution at the final spatial stage
   • Make the prompt encoder generate extra text tokens
   • Remove the need to condition the super-resolution models
   • Reduce sensitivity to domain gaps between cascade stages, which also facilitates parallel training
5. Which capability did Imagen Video notably demonstrate?
   • Rendering legible stylized text and diverse art styles
   • Generating hours of film
   • Real-time gameplay
   • Editing existing photographs only
6. Roughly what output did the full Imagen Video pipeline target?
   • Audio-only clips
   • About 1280x768 at 24 frames per second
   • 8K at 120 fps
   • A single still image
7. Which diffusion parameterization did Imagen Video report using as a design choice?
   • Only a nearest-neighbor color lookup
   • A token-by-token autoregressive text decoder
   • v-parameterization
   • A fixed histogram equalization rule
8. What was progressive distillation used for in Imagen Video?
   • Adding subtitles after generation
   • Replacing the prompt encoder with a database
   • Converting generated video into handwritten text
   • Faster sampling while retaining high-quality generated video