Microsoft and fal partner to expand video generation in Microsoft Foundry, helping developers bring creative ideas into enterprise applications. Video workloads do not all start with the same input or aim for the same result. A creative team might explore several interpretations of a written brief.
A retailer might turn an approved product image into a short motion concept. An application developer might combine image generation and video generation into a workflow that takes an idea from description to frame to scene. Each scenario brings different requirements for visual quality, motion, creative control, turnaround time, and cost.
All of this means that a multi-model solution is necessary to realize a full vision. Today, Microsoft and fal are helping developers find the right tools for their workloads. We are excited to announce that Kling, LTX, and MiniMax video models are rolling out to in public preview in Microsoft Foundry through our new partnership with fal.
Available as a first-party Azure service, these text-to-video and image-to-video options give creators and developers more ways to utilize video generation with the experiences they want to build. The partnership expands the choices available in Microsoft Foundry while bringing video into the broader platform developers use to build AI applications. The goal is to let developers pick the right model for each step of an experience, then connect those steps into a working application.
“fal has seen incredible adoption this year alone with generative media across all industries from advertising to marketing to gaming. Partnering with Microsoft allows both of us to truly enable the wide creative ecosystem and ensure that all AI solutions that lean on generative media, becomes a multi-model one.” – Burkay Gur, Cofounder Bringing fal’s generative media focus to enterprise builders fal is the full stack for generative media, built for creators and developers, with APIs for over 1500 models across image, video, audio, 3D, and generative worlds.
As a truly encapsulating platform, fal provides inference, training, and infrastructure to run generative media workloads in production, while also providing innovative post trained models that pushes the boundaries on speed and quality. That specialization matters because generative media is both a creative and an engineering workload. A compelling demo is only the start.
Production applications also need to handle real traffic, scale with demand, and balance output quality, latency, and cost. Through this partnership, developers will be able to access a selected set of video models in Microsoft Foundry, served by fal and sourced from the companies that build them. The initial portfolio gives teams a choice of models for different creative requirements instead of relying on one model for every task, with more models planned over time.
A portfolio for different video experiences The initial portfolio spans Kling v3 Pro, LTX-2. 5 Fast, LTX-2. 5 Pro, MiniMax H3 Max, and MiniMax H3 Max Turbo.
These models give developers different starting points for generating video from text and animating existing images. A team rapidly exploring creative concepts may prioritize speed and cost, while a team producing fewer, higher-fidelity assets may weigh output quality more heavily. Whether an application starts from text or an existing image, and what it needs in output quality and cost, also shapes the choice.
Kling: From a still image to a directed scene Kling v3 Pro combines cinematic visuals and fluid motion with native audio generation and custom element support. Its image-to-video capabilities give developers control over how a starting image becomes a moving scene, a fit for applications that prioritize visual direction and motion. LTX: One model for fast drafts, one for final output For rapid iteration and previews, LTX-2.
5 Fast is the speed-optimized option; for high-fidelity final output, LTX-2. 5 Pro is the quality-optimized option. Both support text-to-video and image-to-video generation with synchronized audio.
MiniMax: Another approach to video generation MiniMax H3 Max and MiniMax H3 Max Turbo are post-trained variants of MiniMax H3 focused on prompt adherence, aesthetics, and inference performance. For developers, they provide additional options to evaluate against an application’s creative brief, quality requirements, and performance needs. “When building with video models, the details matter: how closely the output follows your prompt, how well it preserves your reference images/videos/audios and how quickly you can try another idea.
Kling gives developers control over detailed reference to video generation. LTX offers fast iteration and high-quality output through its Fast and Pro variants. H3 Max brings fal’s post-training and inference optimizations to improve quality and speed.
Having these models available through Microsoft Foundry gives developers more flexibility to choose what works best for the application they are building.” – Gökay Aydoğan, Creative Engineer Built for enterprise video workloads Bringing video generation into an enterprise application requires more than choosing the right model. Teams need a platform where they can evaluate different options, build with the models that fit their requirements, and operate those experiences alongside the rest of their applications.
The initial portfolio of video models is hosted in the US, and Microsoft Foundry is the first provider to offer a US-hosted Kling endpoint. Key considerations for enterprise teams: Data residency: For organizations with data residency requirements, hosting location is one factor to weigh. Review each model’s documented data processing and storage terms to confirm it meets your requirements.
Model evaluation: Teams can assess quality, performance, and cost, then connect video generation with the other models and capabilities their applications need.
Originally published at techcommunity.microsoft.com
