AI-901 - Implement AI solutions by using Microsoft Foundry (58% of the exam) - Section 2.3
Implement AI solutions with computer vision and image-generation capabilities by using Foundry, including interpreting visual input and creating visual outputs.
Interpreting visual input in prompts by using a deployed multimodal model; creating new visual outputs by using generative image models; and building a lightweight application that includes vision capabilities. The single-best decision is between a multimodal model that interprets an existing image and a generative image model that produces a new one.
lock_openFree sampleImplement AI solutions by using Microsoft Foundrymedium
A logistics firm is comparing two tasks: task one sends a scanned delivery photo to a model and asks "is the parcel damaged?", while task two asks a model to create a new promotional picture of an undamaged parcel from a written brief. Which statement correctly distinguishes the model capabilities these two tasks require?
ATask one needs a multimodal model to interpret the photo, and task two needs an image-generation model to create the promotional picturecheck_circle Correct
BBoth tasks require an image-generation model, because both involve pictures
CBoth tasks require a multimodal model, because it can handle any picture-related workload
DTask one needs an embedding model and task two needs a language model
Interpreting visual input needs a multimodal model, while creating pictures needs an image-generation model. The direction of the visual data decides the capability. When an image goes in and an answer comes out, that is multimodal interpretation; when text goes in and a new picture comes out, that is image generation. The two are not interchangeable.
Why A is correct: Task one supplies an image plus a question and expects reasoning over the visual input, which is multimodal interpretation, while task two turns a text brief into a new picture, which is image generation, so each task maps to a different capability.
Why B is wrong: Sharing the theme of pictures does not make the capabilities identical; task one reads an existing image and must interpret it, which an image-generation model cannot do.
Why C is wrong: A multimodal model interprets visual input but does not synthesise new original pictures, so it cannot satisfy task two despite handling task one.
Why D is wrong: An embedding model only vectorises input and cannot answer whether a parcel is damaged, and a language model outputs text rather than the picture task two needs, so both mappings are wrong.
Exam traps in Implement AI solutions by using Microsoft Foundry
Answers that look right on this material and are not. Each one is a distractor from a different question in the AI-901 bank for this domain.
It is generating a visual output, creating a new picture from the object list.
errorWhy it is wrong: Generating a visual output creates a picture; here the photograph already exists and the model produces text about it, so the direction of the task is reversed.
An image-generation model that turns a text prompt into an original picture
errorWhy it is wrong: Tempting because the scenario involves images, but an image-generation model produces pictures from text; it cannot read a supplied photograph and answer a question about it.
It is image generation, because the application works with product photographs
errorWhy it is wrong: Both workloads involve photographs, which is the trap, but image generation creates a new picture; here an existing photo is analysed and no new image is produced.