Microsoft is bringing five video-generation models from Kling, LTX and MiniMax into Microsoft Foundry through a partnership with generative-media platform fal. The models are rolling out in public preview as a first-party Azure service, giving developers text-to-video and image-to-video options for applications built on Foundry.
The initial lineup comprises Kling v3 Pro, LTX-2.5 Fast, LTX-2.5 Pro, MiniMax H3 Max and MiniMax H3 Max Turbo. Microsoft says the models are intended to give teams choices for different creative and operational needs, rather than requiring one model to serve every stage of a video workflow. More models are planned, according to the announcement.
Video-generation applications can start with a written prompt, an existing image or a sequence of connected model steps. The appropriate choice can depend on factors including motion, visual quality, creative control, turnaround time and cost. The expanded catalog is designed to let developers evaluate options for individual tasks and combine video generation with other models and capabilities in their applications.
Different models for different workflows
Kling v3 Pro is positioned for directed image-to-video work. Microsoft describes it as offering cinematic visuals, fluid motion, native audio generation and support for custom elements. The model also supports text-to-video, according to the pricing details provided in the announcement.
The two LTX models distinguish between rapid iteration and higher-fidelity output. LTX-2.5 Fast is optimized for speed and previews, while LTX-2.5 Pro is presented as the quality-oriented option for final output. Both support text-to-video and image-to-video generation with synchronized audio.
MiniMax H3 Max and H3 Max Turbo are post-trained variants of MiniMax H3. Microsoft says they focus on prompt adherence, aesthetics and inference performance, providing developers with further options to test against an application’s quality and performance needs. H3 Max supports text-to-video, image-to-video and reference-to-video; H3 Max Turbo is listed for text-to-video and image-to-video.
The models are served by fal and sourced from the companies that build them. Fal describes itself as a generative-media platform with APIs for more than 1,500 models across image, video, audio, 3D and generative worlds, as well as inference, training and infrastructure services. The partnership brings a selected group of its video models into Microsoft Foundry for enterprise builders.
Hosting, evaluation and pricing
Microsoft says the initial portfolio is hosted in the United States and describes Foundry as the first provider to offer a US-hosted Kling 3 endpoint. Hosting location is one consideration for organizations with data-residency requirements; the announcement advises customers to review each model’s documented data-processing and storage terms to determine whether a deployment fits their needs.
Teams can assess models for quality, performance and cost before choosing what to use in an application. Microsoft also points to Foundry capabilities for evaluation, optimization, governance, security, observability and infrastructure for supported offerings as teams prepare for production. The availability of these capabilities should be checked for the selected model and deployment.
Rates vary by model and output resolution. The published table gives prices per generated video second for most listed options. MiniMax H3 Max is priced at $0.05 per second at 480p, $0.08 at 768p and $0.16 at 1080p. H3 Max Turbo is $0.025, $0.04 and $0.08 at those respective resolutions.
LTX-2.5 Fast is listed at $0.09 per second at 720p, $0.13 at 1080p, $0.19 at 1440p and $0.30 at 4K. LTX-2.5 Pro is listed at $0.12 per second at 720p and $0.17 at 1080p. For Kling v3 Pro in the US Data Zone, the listed rates are $0.1232 per second with audio off, $0.1848 with audio on and $0.2156 with audio and voice control. Microsoft said pricing would be added to its Azure pricing page in the coming weeks.
Microsoft’s announcement also points developers to explore the models in Foundry and to a quick-start guide for using fal models with ComfyUI. A related Ignite session, BRK256, was scheduled for November 19, 2026, from 11:00 to 11:45 a.m. CT, with speakers from Microsoft and fal.
