Negative Prompts in Image Generation Quiz

Answers and explanations appear as you take the quiz.

1. In classifier-free guidance, what does a negative prompt replace?
   • The positive prompt
   • The random seed
   • The empty unconditional prompt
   • The VAE decoder
2. Why might writing 'no cats' in the main prompt actually produce cats?
   • Text encoders handle negation poorly, so the word 'cats' still pulls the image toward cats
   • The model deliberately ignores instructions it disagrees with
   • The seed already contains a cat
   • Negation automatically doubles the guidance scale
3. Why does classifier-free guidance roughly double the compute per step?
   • It uses two seeds at once
   • It decodes the image twice at the end
   • It upscales the image at each step
   • The model predicts noise for both the positive prompt and the negative or empty prompt
4. According to the guide, which negative term is most likely to have a strong, reliable effect?
   • 'bad anatomy'
   • 'watermark'
   • 'ugly'
   • 'extra fingers'
5. What happens when you use a negative prompt with a guidance-distilled model that skips the unconditional pass?
   • It works twice as strongly as usual
   • It is treated as a second positive prompt
   • It is ignored unless the tool turns true CFG back on
   • It causes the model to crash
6. What is a likely downside of a very long negative prompt list?
   • Less varied images with stiff compositions or distorted colors
   • Faster generation
   • Guaranteed correct hands
   • Higher output resolution
7. In ε = ε_neg + s · (ε_pos − ε_neg), what does s represent?
   • The seed
   • The guidance scale
   • The number of steps
   • The sampler type
8. How is a negative prompt different from a post-processing filter?
   • It is applied only after the final image is produced
   • It removes objects using a hand-drawn mask
   • It affects only the file's metadata
   • It influences every denoising step during generation