What's changed: Added per-section figures (cert-figure-retrofit). Rewrote AI-103 Chapter 4 to align with official skills measured (Apr 16, 2026). Shifted from AI-102-style (prebuilt Vision + Custom Vision) to a generation-centric design: s1 = image/video generation and editing (text→image: GPT-image-1/DALL·E, inpainting = mask editing, reference media, generation/editing controls, text→video: Sora, async); s2 = multimodal understanding (captions/visual QA/alt-text/object & region ID) + Content Understanding visual characteristic extraction (single-task/pro-mode, video analysis) + multimodal Responsible AI (Content Safety image moderation, indirect prompt injection via in-image text, visual policy: watermarks/prohibited symbols/brand)
4.1Image and Video Generation and Editing
In AI-103, computer vision centers on generation more than recognition. Understand—from a developer’s view—deploying image- and video-generation models in Microsoft Foundry, generating images and videos from text prompts and reference media, and editing them via mask-based inpainting and prompt-driven changes.
In AI-102, computer vision centered on recognizing images (classification, object detection, OCR). AI-103 shifts the center to generating and editing images and videos and flexibly understanding them with multimodal models. This section covers generation and editing. You implement it in Microsoft Foundry by selecting image-generation and video-generation models from the catalog, deploying them, and calling them via SDK/REST. Auth uses managed identity (keyless) as in Chapter 1.
4.1.1Image generation (text → image)
Image generation creates new images from a text prompt (e.g., "a dog walking on a beach at dusk, watercolor style"). In Foundry you deploy and use image-generation models such as GPT-image-1 or DALL·E. Beyond the prompt, you can pass reference media (a reference image) to guide style or composition. You tune output with generation controls (image size, quality, number of images n, etc.). The key difference from pre-AI-102 vision is that you create images anew rather than recognize existing ones.
4.1.2Image editing (inpainting and masks)
Generation models also edit existing images. The flagship is inpainting: specify a region with a mask and redraw only the masked region per the prompt (e.g., "replace just the background behind the person with a forest"). Without a mask, you can apply prompt-driven modifications to the whole image. These are controlled by editing controls (which region, how strongly to change it). Distinguish generation (create new) from editing (change an existing image), and remember that in editing the mask = the region to change.
4.1.3Video generation and editing
AI-103 also covers video generation. Deploy a video-generation model (text→video, e.g., Sora) in Foundry and generate short videos from text prompts and reference media. You can edit generated videos—adjusting length, replacing segments, and so on. Because video costs more compute and time than images, implement it assuming asynchronous generation with progress checks (consistent with Chapter 3’s messaging/long-running patterns).
| Goal | Approach | How it’s done in Foundry |
|---|---|---|
| Create an image from text | Image generation | Deploy GPT-image-1 / DALL·E and send a prompt |
| Redraw just part of an image | Inpainting (mask) | Pass image + mask + prompt to the edit API |
| Guide style with a reference image | Reference media | Pass a reference image alongside the prompt |
| Create a short video from text | Video generation | Call a video model (e.g., Sora) asynchronously |
Common: (1) "create a new image from text" = image-generation model (GPT-image-1 / DALL·E). (2) "redraw only part of an image via prompt" = inpainting = specify the region with a mask. (3) "guide style/composition with a reference image" = reference media. (4) "text → video" = video-generation model (e.g., Sora). (5) Distinguish generation (create new) from editing (change existing).
Watch out: (1) generation (make images) vs recognition (classify/detect existing images) are different goals—AI-103 CV is generation-centric. (2) inpainting = redraw the masked region (vs whole-image edits without a mask). (3) image and video generation are separate deployments. (4) Video generation is slow—don’t block synchronously (use async + progress checks).
4.1.4Section summary
- AI-103 CV is generation-centric: deploy and call image/video generation models in Foundry
- Image generation = text → image (guide with reference media; tune size/quality/count)
- Editing = inpainting (mask the region and redraw) / prompt-driven modifications
- Video generation = text → video (e.g., Sora); slow, so implement async + progress checks
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. How do you create a new image from a text prompt in Foundry?
Q2. Which technique masks part of an image and redraws that region per a prompt?
Q3. Which model do you use to generate a short video from a text prompt?
Q4. In image generation, what is passing a reference image to guide style/composition called?
Keep track of your progress
The full study guide is free to read. Sign up free to practice with the question bank, track what you have read, review your mistakes, and highlight passages.