Five fal-served video models — Kling v3 Pro, LTX-2.5 Fast, LTX-2.5 Pro, MiniMax H3 Max, and MiniMax H3 Max Turbo — added to Microsoft Foundry in public preview on November 5, 2026, most with text-to-video, image-to-video, and native or synchronized audio.
Last verified · 2026-11-07 · by Moe Ameen
On November 5, 2026, Microsoft brought five AI video models served by fal into Microsoft Foundry, its platform for building and running AI applications on Azure. The lineup, in public preview, is Kling v3 Pro, LTX-2.5 Fast, LTX-2.5 Pro, MiniMax H3 Max, and MiniMax H3 Max Turbo. fal — which Microsoft describes as "the full stack for generative media," with APIs to more than 1,500 models — supplies the serving layer; Foundry wraps it in Azure identity, billing, and governance. This is a curated selection of fal's catalog, and Microsoft says more models are planned.
The five cover the common generation modes, and audio is built in rather than added afterward. Kling v3 Pro is the cinematic option — fluid motion, native audio, and custom-element support — running on a US Data Zone deployment Microsoft calls the first US-hosted Kling 3 endpoint. LTX-2.5 comes in two flavors: Fast, tuned for rapid iteration, and Pro, tuned for high-fidelity final output; both do text-to-video and image-to-video with synchronized audio. MiniMax H3 Max and H3 Max Turbo are post-trained variants of MiniMax H3 focused on prompt adherence, aesthetics, and inference performance, with H3 Max also listing reference-to-video.
Two things to keep in mind before you build on it. It is a public preview, not general availability, so access, pricing, and region coverage are still moving — Microsoft says per-model pricing lands on the Azure pricing page in the coming weeks, and the published unit has been inconsistent (per video second in one place, per minute in another). And the initial models are US-hosted only, which matters if you have data-residency requirements outside the US.
The boundary to design around is the shape of the output. Foundry is a model-access platform: you call an endpoint and get back a raw video file. It does not detect clips, style captions, reframe to a platform's aspect ratio, generate carousels or blogs, or schedule and publish anything. Those are a separate job — and the layer where a generated clip becomes finished content.
Here is the honest gap, and the good news inside it: Foundry gives you a fantastic raw clip and then stops. You cannot post a bare MP4 — it has no captions, it is the wrong aspect ratio for four of your five feeds, and it is one asset when a week of content needs ten. Using Foundry at all also means an Azure deployment and API calls, which is fine for a developer and a wall for everyone else. [Kompozy](/) is the no-code layer that starts exactly where the endpoint ends, and it doesn't make you wire up a model at all.
Picture a real week. You generate a cinematic 8-second hook from Kling v3 Pro and a couple of quick scene shots from LTX-2.5 Fast, download them from Foundry, and drop them into Kompozy as sources. From there one clip becomes a feed's worth of content: Kompozy reframes each shot to vertical, square, and wide, burns word-synced captions in your brand's style, and — if you stitched a longer cut — runs it through [clipping](/glossary/clipped-short) to pull several shorts. Then it generates the formats no video endpoint produces: brand-exact [Carousel Posts](/glossary/hyperframes) and quote cards through HyperFrames, per-platform text, a blog article, an email newsletter, and an avatar-narrated [Persona Short](/glossary/persona-shorts) that needs no camera — all held to one [Persona Brief](/glossary/persona-brief) so the whole set sounds like one brand. [Autopilot](/glossary/autopilot) schedules and publishes it across the eight social platforms plus blog and email behind a per-post review gate. And because Kompozy already bundles its own video stack — HeyGen avatars, fal-generated VFX hooks, Pexels b-roll — a Foundry clip slots in as an optional premium source rather than a dependency. See the honest [review](/reviews/microsoft-foundry-fal-video) and the [platform-vs-product alternative](/alternatives/microsoft-foundry-fal-video) for the full picture.
Five, in public preview as of November 5, 2026: Kling v3 Pro, LTX-2.5 Fast, LTX-2.5 Pro, MiniMax H3 Max, and MiniMax H3 Max Turbo. They are served by fal, Microsoft says more are planned, and most support text-to-video and image-to-video with native or synchronized audio.
Kling v3 Pro is the cinematic pick (fluid motion, native audio, custom elements); LTX-2.5 Fast favors speed for drafts and LTX-2.5 Pro favors fidelity for finals; MiniMax H3 Max and Max Turbo are post-trained for prompt adherence and performance, with H3 Max adding reference-to-video. Choose on whether you need quality, speed, or control.
Effectively, yes. Foundry is a model-access platform on Azure — using the video models means deploying them and calling an API, with no creator-facing interface. If you want to generate and post content without that, pair the output with a no-code content engine like Kompozy, or use Kompozy's own bundled video generation.
Foundry returns a raw video file; it does not caption, reframe, schedule, or publish. Bring the clip into Kompozy, which clips long footage, burns on-brand captions, reframes per platform, generates carousels, a blog, and a newsletter from the same source, and auto-publishes across the eight social platforms plus blog and email.
No. It launched in public preview, the initial models are US-hosted, and per-model pricing is still being published on the Azure pricing page. Confirm availability, region, and rates in the Foundry portal before building a production pipeline on it.