AI-901: Create new visual outputs by using generative models
For AI-901, this means understanding how Microsoft Foundry can use generative AI models to create images or visual content from prompts.
Microsoft lists this skill under “Implement AI solutions with computer vision and image-generation capabilities by using Foundry” in the AI-901 study guide.
What you need to know
A generative image model creates new visual content based on a prompt, for example:
“Create an image of a modern data platform architecture using Microsoft Fabric, Azure, and Power BI.”
The model does not simply retrieve an existing image. It generates a new visual output based on patterns it learned during training.
Typical visual generation tasks
You should recognise scenarios such as:
| Requirement | Best fit |
|---|---|
| Create an image from a text prompt | Image generation model |
| Create product mock-ups | Generative image model |
| Produce marketing visuals | Generative image model |
| Generate diagrams or concept art | Generative image model |
| Edit or vary an existing image | Image editing / image variation capability |
| Describe what is in an image | Multimodal vision model, not image generation |
In Foundry
In Microsoft Foundry, you would typically:
- Choose a suitable image-generation model
- Provide a clear text prompt
- Optionally specify style, format, size, or quality
- Generate the image
- Review the output for accuracy, safety, and suitability
Example prompt
Create a professional illustration of a cloud-based analytics platform with data flowing from source systems into a lakehouse, then into dashboards.Exam-style distinction
For AI-901, remember:
| Task | Model type |
|---|---|
| “What is shown in this picture?” | Vision / multimodal model |
| “Extract text from this image” | OCR / information extraction |
| “Generate a new image from this description” | Generative image model |
| “Create a visual design from a prompt” | Generative image model |
Key exam point
Generative image models create new visual outputs.
Computer vision models interpret existing visual inputs.