IP-Adapter for Image Prompts Quiz

Answers and explanations appear as you take the quiz.

1. What core problem does IP-Adapter address?
   • Diffusion models are too slow to run on consumer GPUs
   • Images cannot be encoded by neural networks
   • Text prompts struggle to describe specific visual details like a face or style
   • Text prompts are too short to fit in memory
2. How does the original IP-Adapter add image conditioning to text-to-image diffusion?
   • It retrains every parameter of the base model
   • It concatenates raw image pixels into the text tokenizer
   • It removes text conditioning whenever an image is supplied
   • It adds a separate image cross-attention path alongside text attention
3. Which checkpoint family did the original IP-Adapter paper demonstrate reusing a trained adapter with?
   • Any diffusion architecture regardless of its base model
   • Compatible custom checkpoints fine-tuned from the same base diffusion model
   • Only checkpoints with all adapter parameters retrained from scratch
   • Only a text encoder without a diffusion network
4. Roughly how many trainable parameters does the original IP-Adapter add?
   • About 22 million
   • About 7 billion
   • About 500
   • About 100 billion
5. Which tool is IP-Adapter commonly combined with for layout control?
   • A spell-checker
   • A spreadsheet
   • A text summarizer
   • ControlNet
6. What encodes the reference image before its features enter the model?
   • A random number generator
   • The text tokenizer
   • A frozen CLIP image encoder
   • A database lookup
7. After the CLIP image encoder, what job does IP-Adapter's trainable projection perform?
   • It changes the diffusion model's tokenizer vocabulary
   • It maps the reference embedding into features compatible with the attention layers
   • It converts the generated image into a segmentation label
   • It trains CLIP from scratch for each reference photo
8. In IP-Adapter's decoupled attention design, how are text and image attention contributions used together?
   • Their attention outputs are combined to condition generation
   • Text attention is permanently discarded after training
   • Image features are used only to rename the output file
   • Each contribution must generate a separate finished image