Microsoft has added five AI video-generation models to Microsoft Foundry through a new partnership with fal, giving developers text-to-video and image-to-video tools inside Azure’s AI application platform. The lineup — Kling v3 Pro, LTX-2.5 Fast, LTX-2.5 Pro, MiniMax H3 Max and MiniMax H3 Max Turbo — began rolling out in public preview on 5 November as a first-party Azure service, with the models hosted in the United States.
The five models
Kling v3 Pro is the cinematic option, combining motion with audio generation and element support for directing a scene from an image. Its maker banked nearly $3 billion in a funding round days earlier, with Kuaishou keeping control. Microsoft lists the model under its US Data Zone, and says Foundry is the first provider to offer a US-hosted Kling endpoint.
LTX-2.5 comes in two versions built for different stages of a project. The Fast variant is tuned for rapid drafts and previews, while the Pro variant targets higher-fidelity final output. Both support text-to-video and image-to-video generation with synchronised audio, so a team can sketch an idea quickly and then render the finished version without switching platforms.
MiniMax H3 Max and MiniMax H3 Max Turbo are post-trained variants of MiniMax H3, focused on prompt adherence, aesthetics and inference speed. The Max version adds reference-to-video generation, letting developers steer a clip from reference images or video, while the Turbo version trades some quality for faster, cheaper output.
Pricing is charged per second of generated video and varies by model, resolution and audio settings. Kling v3 Pro runs from $0.1232 per second with audio off to $0.2156 with voice control, while MiniMax H3 Max Turbo starts at $0.025 per second at 480p. Microsoft says the prices will be added to its Azure pricing page in the coming weeks.
Our Read
Foundry is becoming a one-stop shop for multimodal building, and Microsoft is renting the video models rather than building them. Fal offers APIs for more than 1,500 models, but Microsoft has curated just five. The contrast with rivals is distribution: Google and OpenAI sell their own video models, while Microsoft sells a shelf of other makers’ models inside Azure.
Microsoft has also been pruning third-party models from Copilot Studio even as it adds video models to Foundry, so the curation cuts both ways. The caution flags are plain: this is a public preview with no general-availability date, and no pricing has reached the Azure pricing page yet.
Microsoft’s own post is inconsistent on pricing, calling the rates per minute in its text while its table prices them per second. Developers should confirm the billing unit on the Azure pricing page before budgeting for production use at scale.
Why It Matters
Developers can now compare five video models inside one platform instead of juggling vendors, accounts and bills. A US-hosted Kling endpoint answers data-residency worries that have kept teams from using Chinese models. Video generation is moving from creative tools into workflows, and fal — which doubled to a $16 billion valuation earlier — is the infrastructure bet behind that shift.
Background
Foundry is Microsoft’s platform for building AI applications, and video now joins the image models already offered there. Microsoft opened its Copilot harness to Foundry builders in October, and the fal deal lands days after fal’s own funding news. Microsoft will present the partnership in an Ignite session on 19 November.
Source: Microsoft’s announcement.
0 Comments