Why AI Image Generators Get Hands Wrong Quiz

Answers and explanations appear as you take the quiz.

1. According to the guide, what does a diffusion model mainly learn from image-caption pairs?
   • Local visual patterns such as texture and lighting
   • Explicit anatomical rules about fingers
   • A built-in 3D skeleton of the human body
   • A counting procedure for body parts
2. How does latent compression in Stable Diffusion make hands harder to draw?
   • It removes color information
   • It deletes hands from the training images
   • It shrinks the image by a factor of eight in each dimension, leaving a hand only a few latent cells
   • It converts the image into text
3. Why do captions give models little help with hands?
   • Captions are written in an obscure language
   • Captions rarely mention hands or describe their pose
   • Captions are removed before training
   • Captions describe only hands and nothing else
4. Which Midjourney version does the guide say was widely noted for better hands in 2023?
   • Version 1
   • Version 2
   • Version 3
   • Version 5
5. What does an ADetailer-style detect-and-repaint workflow do?
   • Finds hand regions, regenerates each one at higher effective resolution through inpainting, and blends it back
   • Adds a watermark over the hands
   • Replaces hands with gloves automatically
   • Deletes the hands from the image
6. Why do negative prompts like 'extra fingers' usually help little?
   • They are too long for the encoder
   • They only work on text rendering
   • They switch off classifier-free guidance
   • Few training captions described those errors, so the model has only a weak concept to steer away from
7. What does a ControlNet OpenPose input provide when fixing hands?
   • A color palette
   • A new random seed
   • A skeleton with hand keypoints that supplies the geometry
   • A longer caption
8. A suspected fake image has anatomically perfect hands. What should a fact-checker conclude?
   • The image is definitely a real photograph
   • Correct hands prove nothing either way, because newer generators often draw them correctly
   • The image must come from Midjourney
   • The image must carry a watermark