InstructPix2Pix Instruction Editing Quiz

Answers and explanations appear as you take the quiz.

1. How do you tell InstructPix2Pix what to change?
   • By writing a plain-language editing instruction
   • By drawing a precise mask over the region
   • By uploading a second reference image
   • By adjusting sliders for each pixel
2. How was InstructPix2Pix's training data created?
   • By hiring humans to edit millions of photos
   • Using GPT-3 for caption pairs and Prompt-to-Prompt to make image pairs
   • By scraping before/after photos from social media
   • By manually masking each training image
3. What do InstructPix2Pix's two guidance scales control?
   • Resolution and color depth
   • Training speed and inference speed
   • Adherence to the instruction versus fidelity to the original image
   • Number of objects and number of colors
4. Why does the model tend to leave unmentioned parts of the image alone?
   • Because it only edits the image center
   • Because it ignores the input image entirely
   • Because it always uses a mask internally
   • Because instructions describe a change rather than a whole new scene
5. Does InstructPix2Pix require separate per-image fine-tuning or inversion before applying an edit?
   • No; the trained model can apply an instruction without per-example fine-tuning or inversion
   • Yes; every image must be used to train a new diffusion model
   • Yes; an artist must invert each pixel by hand
   • No; it edits images without running any diffusion steps
6. If an edit is too aggressive and loses the original subject, what helps?
   • Lowering the resolution
   • Raising the image-fidelity guidance scale
   • Removing the instruction text
   • Switching to grayscale
7. What is one limitation to check after an InstructPix2Pix edit?
   • The model always preserves every unmentioned pixel exactly
   • The output is guaranteed to match historical truth
   • The model may change details beyond the intended edit
   • Every instruction must be converted into a hand-drawn mask
8. What generative model provided the image prior used in InstructPix2Pix’s synthetic training-data pipeline?
   • AlphaFold
   • Whisper
   • BERT
   • Stable Diffusion