Make-A-Video Text-to-Video Quiz

Answers and explanations appear as you take the quiz.

1. What was the central innovation that let Make-A-Video avoid needing labeled text-video pairs?
   • It learned appearance from captioned images and motion from unlabeled video separately
   • It manually labeled millions of videos with captions
   • It only generated videos from existing video clips
   • It used a single dataset of paired text and video
2. Which organization released Make-A-Video and when?
   • Google in 2021
   • Meta AI in 2022
   • OpenAI in 2023
   • Stability AI in 2024
3. How does Make-A-Video add motion capability to an image model?
   • By replacing all spatial layers with audio layers
   • By rendering each frame with a game engine
   • By adding temporal (pseudo-3D) layers on top of pretrained spatial layers
   • By averaging many random images together
4. What do the frame-interpolation and super-resolution stages do?
   • They translate the prompt into another language
   • They remove all motion from the clip
   • They compress the video for storage
   • They add more frames and increase spatial resolution
5. What was a typical limitation of Make-A-Video's output?
   • Short, low-resolution clips with flicker and warping
   • Hours-long flawless films
   • Perfectly stable 4K video
   • Audio-only output
6. Besides text prompts, what visual input can Make-A-Video use to create a video?
   • Only a waveform
   • An image
   • A spreadsheet formula
   • A source-code repository
7. What can Make-A-Video do with an existing video, according to Meta’s announcement?
   • Recover the original camera’s sensor data
   • Guarantee an exact copy of every frame
   • Create a new video similar to it
   • Convert the clip into a signed legal transcript
8. What did Meta say about the datasets used in Make-A-Video research?
   • Every training example was privately collected by Meta
   • The model was trained only on generated video
   • It used no video data
   • The system used publicly available datasets