Stable Video Diffusion Quiz

Answers and explanations appear as you take the quiz.

1. What input does a public Stable Video Diffusion image-to-video checkpoint use?
   • A single still image
   • A 3D mesh file
   • A full text screenplay only
   • A microphone recording
2. What did SVD add to the still-image Stable Diffusion architecture to handle motion?
   • A larger color palette
   • Temporal layers connecting frames over time
   • Audio synthesis modules
   • A spell-checker for captions
3. Stable Video Diffusion performs its denoising in what kind of space to save compute?
   • Raw full-resolution pixel space
   • The frequency domain only
   • A compressed latent space
   • A text token space
4. What was significant about SVD’s research release?
   • It removed all license conditions
   • It was accessible only as a closed cloud API
   • It released audio weights but no video model
   • It made code and model weights available for research and adaptation under their license terms
5. Roughly how many frames do SVD's public checkpoints typically generate?
   • Around 14 to 25 frames
   • Exactly 1 frame
   • Several thousand frames
   • An unlimited continuous stream
6. What training approach did Stability AI emphasize for SVD?
   • A single one-shot training run on random clips
   • A staged recipe: image pretrain, video pretrain, then high-quality finetune
   • Training only on still images
   • No pretraining at all
7. What downstream video adaptation does the SVD paper discuss using LoRA?
   • Speaker-identity adaptation for audio
   • Automatic SQL generation
   • Camera-motion-specific adaptation
   • Dependency-label prediction
8. What additional visual-generation capability did the SVD paper explore from its base model?
   • Generating a dependency parse for every frame
   • Separating a song into instrument stems
   • Estimating speech intelligibility
   • Fine-tuning a model to jointly generate multiple views of objects