Build On-Device AI Experiences w/ Windows AI APIs | THR888
Greg Woo and Savo Icagic present practical APIs, patterns, and tools for adding local AI to Windows applications, with demonstrations of language-model features and video enhancement.
Windows AI APIs in action
This advanced Windows session from Microsoft Ignite 2026 introduces Microsoft Foundry on Windows and the Windows AI API catalog. It covers text, image, speech, and video scenarios, then demonstrates how applications can use local models for language tasks and vision features.
Why run AI locally?
The session highlights four benefits of on-device processing:
- Privacy: Processing can take place on the PC rather than sending data to a remote service.
- Low latency: Local execution supports responsive application experiences.
- Offline use: AI features can work without a network connection.
- Bundled APIs: The presenters discuss free bundled APIs for adding AI capabilities to Windows apps.
Language capabilities and customization
The presenters introduce the Instruct small language model and its performance improvements, followed by live demonstrations of:
- Text generation for producing new text.
- Rewriting for revising existing content.
- Summarization for condensing longer text.
- Structured output for generating JSON that follows a supplied schema.
A partner showcase with Flux Ink demonstrates handwriting AI and next-word prediction. The session also covers LoRA fine-tuning for model customization, including a Microsoft Education use case.
Video Super Resolution
The closing demonstrations focus on Video Super Resolution and partner integrations, showing another application of Windows AI APIs beyond language processing.
Session chapters
- 00:00 — Introduction to Windows AI APIs and local AI on PCs
- 01:17 — Privacy, low latency, offline use, and free bundled APIs
- 02:29 — Microsoft Foundry on Windows and Windows AI APIs
- 03:03 — API catalog for text, image, speech, and video scenarios
- 03:56 — Instruct small language model and performance improvements
- 05:02 — Text generation, rewriting, and summarization demonstrations
- 08:23 — Flux Ink handwriting AI and next-word prediction showcase
- 09:14 — Structured output and JSON schema demonstration
- 11:50 — LoRA fine-tuning and the Microsoft Education use case
- 14:44 — Video Super Resolution, partner integrations, and closing resources