AI
Sora on Azure OpenAI: API Shape and Production Use Cases
Sora is now in public preview on Azure OpenAI. The API is job-based: submit a text prompt with resolution, duration, and variation controls to the video generation endpoint, poll by job ID, and retrieve by generation ID. It produces up to 20 seconds of 1080p video per generation. Don Scott demonstrated three scenarios: a luxury fragrance commercial, a manufacturing safety training clip showing workers in protective gear, and a film studio storyboard of a monster discovering a typewriter. Each ran in minutes.
Healow Genie: AI Contact Center for Healthcare
Healow’s Genie contact center handles appointment bookings, refill requests, billing questions, and referral queries across 16,000 physician practices and 180,000 providers. Doctor’s offices receive up to 15 million patient communications per month; Genie takes inbound calls, texts, and chat without a human agent, and routes after-hours calls to the on-call provider.
“It saves a lot of time for practice staff.” — Sidd Shah
Separately, Healow’s Sunoh ambient listening product transcribes doctor-patient conversations into structured clinical notes automatically. Image AI routes incoming faxes, splits multi-patient documents by page, and files each page into the correct patient record and folder.
GPT-image-1 and Sora: Model Capabilities on Azure AI Foundry
GPT-image-1 reached API availability less than a month before this session. It accepts both text and image as input, supports inpainting and text rendering at 1024+ pixels, and improves on DALL-E for prompt adherence. Sora, confirmed for public preview at the same event, generates up to 20 seconds of 1080p video from text.
“It’s a flexible, high-fidelity video generation model,” — Malavika Pande
Both run on Azure’s enterprise infrastructure: C2PA provenance, abuse monitoring, billing, and identity come included. The images playground 2.0 also supports Stable Diffusion 3.5 Large, Stable Image Ultra, and Flux 1.1 Pro from Black Forest Labs.
Azure AI Foundry Playground: From Prompt to Production Code
Thasmika Gokal walked through the Sora video playground demo (26:48) live, iterating a two-word “solar system” prompt through aspect ratio, resolution, and duration controls until the output matched her e-learning use case. The images playground 2.0, released 20 days before the session, exports contextual code samples in Python, JavaScript, C#, Go, or cURL. The Open in VS Code demo (36:00) showed one-click import of the API key, endpoint, and code into VS Code for the web, with automatic environment setup, a requirements.txt, and a GitHub clone path. No manual credential handling required.
WPP and T&Pm: Multimodal Ad Production at Scale
T&Pm, a WPP agency, built two Sora-powered pipelines. The WPP comic book pipeline demo (50:42) shows a Python script generation tool trained on comic book structure that outputs a seven-box-grid script, sends it to the Sora API, and uses GPT-4o twice: before generation to verify brand tone and format, and after to flag unsafe or off-brief output. The second pipeline, the meal machine, automates plated food shots for consumer packaged goods brands. A prompt engine reads dish name, camera angle, and specs, then generates several-thousand-character Sora prompts in seconds.
“There’s some human crafting to be done.” — Cameron Milne
Notable Quotes
It saves a lot of time for practice staff. Sidd Shah · ▶ 10:56
It’s a flexible, high-fidelity video generation model, Malavika Pande · ▶ 17:59
There’s some human crafting to be done. Cameron Milne · ▶ 49:10
Key Takeaways
- Sora’s job-based API generates up to 20 seconds of 1080p video from text prompts, available now on Azure OpenAI.
- Healow’s Genie contact center handles 15 million monthly patient communications across 16,000 physician practices using voice, chat, and ambient listening AI.
- Azure AI Foundry’s Open in VS Code feature moves a model’s API key, endpoint, and code sample into a working VS Code project in one click.