← AI-103: AI apps and agents on Azure
04 / 6 · 40 MIN

Visual and multimodal content

Generate, edit, and interpret media with explicit quality and lifecycle criteria.

Concept and mechanism

Visual generation and visual understanding solve different tasks. To edit an area, use a reference image and compatible mask when supported by the model and API; compare the result, including areas intended to stay unchanged. DALL-E 3 retired on Azure in March 2026, so current examples should check supported image models. The Sora 2 example is in preview. Generated video is asynchronous work: retain its identifier, track state, and retrieve the artifact. Remix depends on the contract and references accepted by the API. A temporary link is not a durable archive. Retain version, prompt, and provenance under retention policy and review material before training use.

Guided application

To understand a fictional diagram, request answers tied to visual evidence and flag unreadable elements. Alternative text should communicate image function; complex diagrams can need detailed relationship descriptions. Evaluate adversarial visual content, including embedded text attempting to change instructions, and retain tool boundaries. The AI-103 syllabus mentions Content Understanding pro mode, but the old API retired. The agentic 2026-06-01-preview alternative is not a drop-in replacement: the initial preview accepts one file per document analysis and requires configuration and schema review. Adoption should consider limitations, latency, and cost; an exam-syllabus reference does not establish current operational support.

IN PRACTICE

A queued video is not ready. Reopening the page should resume tracking the same work without automatically generating another.

Common pitfalls

Mask as an absolute guarantee; temporary URL as archive; OCR as instruction authority; exam-mentioned mode as a current API.

Related topics: Architecture, project, and release · Generative applications and evaluation · Agents, memory, and approvals

Take this idea with you

Validate artifacts and visual evidence against the actually supported contract and lifecycle.

Create account

Reference: Current image generation editing and DALL-E3 retirement · AI-103 skills measured2026-04-16; Microsoft Foundry documentation2026-09-30