AIF-C01 Practice Question 23: Image Generation Model Choice
AWS Certified AI Practitioner (AIF-C01) · Test 3 · question 3 of 10
A marketing team writes short descriptions of product concepts, such as "a matte black water bottle on a sunlit kitchen counter", and wants a generative AI service to produce original images from those descriptions for early design reviews. They have been using a text-only foundation model, which only replies with written descriptions of what such a picture would look like. Which type of model do they actually need?
Show answer and explanation
Answer:
The requirement is text in, image out. That is text-to-image generation, the job of an image generation model such as Amazon Nova Canvas on Amazon Bedrock, which generates an image capturing the concepts described by the prompt (option A). Models of this kind are commonly built on diffusion architectures, which start from noise and repeatedly denoise it toward an image that matches the prompt. Choosing a model starts with matching its input and output modalities to the task, and a text-only model can never emit an image no matter how the prompt is written.
- B is wrong because it runs the pipeline backwards: it accepts an image and produces text. That is useful for captioning or visual question answering, not for creating new artwork. - C is wrong because an embedding model outputs a numeric vector, not an image. Embeddings power semantic search and retrieval; they could find an existing photo, but not generate one. - D is wrong because a speech model's output modality is audio. Narration does not help a team that needs to look at a proposed product design.
Reference: Generating images with Amazon Nova Canvas