LoRA Sliders for Image Editing Quiz

Answers and explanations appear as you take the quiz.

1. What does the 'Low-Rank' in LoRA refer to?
   • Using two small matrices to approximate weight changes instead of editing all weights
   • Lowering the resolution of the output image
   • Reducing the number of diffusion steps
   • Ranking concepts from least to most important
2. Why can you stack multiple LoRA sliders at once?
   • Because they overwrite the base model in sequence
   • Because each is a small, separate module added to the frozen base model
   • Because the model retrains between each slider
   • Because only one slider is ever active at a time
3. What does increasing a slider's strength value typically do?
   • Adds random noise to the image
   • Switches to a different base model
   • Pushes the targeted concept more strongly in the positive direction
   • Lowers the file size of the LoRA
4. Compared to wrestling with text prompts, a key advantage of sliders is:
   • They require no base diffusion model
   • They eliminate the need for any prompt
   • They produce only black-and-white images
   • Continuous, repeatable control over a single attribute
5. How many unique Concept Sliders does the paper report composing without degrading output quality?
   • More than 50 in the reported composition experiment
   • Exactly two, with no larger compositions tested
   • One slider per image, with no stacking
   • An unlimited number for every model
6. What concept did the Concept Sliders work introduce?
   • A way to compress images with diffusion
   • Training LoRA directions that map to a single editable attribute
   • A new sampler for faster generation
   • A method to caption images automatically
7. What training examples can Concept Sliders use to define a visual attribute direction?
   • Only a complete retraining of the diffusion model
   • A list of file names with no examples
   • A small set of text prompts or sample images
   • A video transcript unrelated to the target attribute
8. Why does Concept Sliders seek to minimize interference with unrelated attributes?
   • So the frozen base model must be retrained for every edit
   • So all sliders produce the same visual change
   • So a slider cannot be combined with another module
   • So changing one slider is less likely to alter other aspects of the image