What happened
Microsoft announced that Microsoft Foundry now offers five AI video generation models through a partnership with Fal. The models, which began rolling out in public preview on November 5, 2026, include Kling v3 Pro, two variants of LTX-2.5, and two variants of MiniMax H3 Max. These models support text-to-video, image-to-video, and audio-enabled generation. Microsoft states that Fal provides the , training, and infrastructure for these generative media workloads, bringing them into Foundry as a first-party Azure service. The announcement includes a pricing table with rates in US dollars per generated video second, though Microsoft notes that final pricing details will be added to the Azure pricing page in the coming weeks. Deployment classifications vary, with Kling v3 Pro listed under US Data Zone and the LTX and MiniMax models listed under Global Standard.
Microsoft Foundry has integrated five AI video generation models through a partnership with Fal, marking a significant expansion of its generative media capabilities. The models, which entered public preview on November 5, 2026, include Kling v3 Pro, two variants of LTX-2.5, and two variants of MiniMax H3 Max. These models support various generation methods, including text-to-video, image-to-video, and audio-enabled generation, allowing developers to choose the approach that best fits their specific project needs.
The partnership positions these models as a first-party Azure service, meaning they are served by Fal but accessed through Microsoft’s platform. Microsoft states that Fal provides the underlying , training, and infrastructure for these workloads. This integration allows businesses to evaluate different models within the same environment they use to develop other AI applications, potentially streamlining the development process and reducing the need to manage multiple external APIs.
Pricing is a critical component of this launch, with rates varying based on the selected model, output resolution, and audio settings. The source notes a discrepancy in Microsoft’s announcement, where introductory text refers to rates per minute while the pricing table specifies US dollars per generated video second. The article follows the table’s per-second unit, but advises developers to confirm the applicable billing unit through Azure documentation before estimating production budgets. Microsoft has indicated that these prices will be formally added to its Azure pricing page in the coming weeks.
Why it matters
This expansion allows developers to evaluate multiple specialized video models within a single Azure platform, rather than relying on a single generation system. This flexibility is significant for businesses with varying requirements for output quality, generation speed, audio capabilities, and cost. For example, marketing teams can use faster, lower-cost models for concept exploration and higher-quality models for final output. The integration also simplifies workflow management by combining video generation with other AI capabilities within the same development environment. However, the public preview status means that pricing and availability are not yet stable, and developers must verify specific deployment terms, particularly regarding data residency, as the models have different zones.
The primary value of this update lies in the ability to compare specialized models within a single development platform. Video projects often have diverse requirements, and having access to multiple models allows developers to match the tool to the task. For instance, a marketing team might use a faster, lower-cost model to generate numerous draft concepts before selecting a higher-quality model for the final production output. This flexibility can lead to more efficient workflows and potentially lower costs for iterative processes.
The distinction in deployment classifications is also significant for enterprise users. Microsoft’s pricing information lists Kling v3 Pro under the US Data Zone, while the LTX-2.5 and MiniMax models are listed under Global Standard. This difference is crucial for businesses with data residency or compliance requirements, as it implies that not all models offer identical data handling guarantees. Developers must review the specific terms for each model to ensure they meet their organizational compliance standards.
Furthermore, the public preview status indicates that while the models are available for testing, they are not yet guaranteed for stable production use. Developers should be cautious about building critical production workflows around these services until general availability is announced and pricing is finalized. The ability to combine video generation with other AI capabilities in larger workflows represents a practical implication for developers looking to create more complex, multi-modal applications.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
crm_get_transaction(id='4092').A route planner searches possible journeys using explicit rules. What does this illustrate about AI?
What to watch next
Developers should monitor the official Azure pricing page for the final confirmation of billing units and rates, as the current announcement contains inconsistent wording regarding per-minute versus per-second pricing. Additionally, businesses with strict data residency requirements need to review the specific processing and storage terms for each model, as the US Data Zone and Global Standard classifications imply different compliance implications. The timeline for general availability and the introduction of additional models through the Fal partnership are also key factors to watch.
The most immediate item to watch is the resolution of the pricing inconsistency. Developers need to confirm whether the billing unit is per second or per minute, as this significantly impacts cost estimation for large-scale video generation. The official Azure pricing page will be the definitive source for this information once updated.
Businesses with strict data residency requirements should closely monitor the deployment terms for each model. The difference between US Data Zone and Global Standard classifications means that compliance obligations may vary. It is essential to verify how data is processed and stored for each specific model to avoid potential regulatory issues.
The timeline for general availability is another key factor. Microsoft has not announced a specific date for when these models will move from public preview to general availability. Until then, developers should assume that pricing, availability, and supported capabilities may change. Additionally, Microsoft has stated that it plans to introduce more models through the Fal partnership over time, so the current list of five models is likely to expand.