Video models
Kling, LTX and MiniMax video models arrive in Microsoft Foundry
Microsoft and fal have put Kling, LTX and MiniMax video models into Microsoft Foundry in public preview, as a first-party Azure service. Developers get text-to-video and image-to-video next to the rest of Azure's AI tools.
Microsoft announced on 5 November 2026 that Kling, LTX and MiniMax video models are rolling out in public preview in Microsoft Foundry, its platform for building AI applications. The models arrive through a new partnership with the generative-media platform fal, and Microsoft says they are offered as a first-party Azure service.
What is available
The first batch is five models, all offering text-to-video and image-to-video:
- Kling v3 Pro, which Microsoft says pairs cinematic visuals and fluid motion with native audio generation and detailed control over how a reference image becomes a moving scene.
- LTX-2.5 Fast and LTX-2.5 Pro, a speed-optimised model for drafts and previews and a quality-optimised one for final output, both with synchronised audio.
- MiniMax H3 Max and H3 Max Turbo, variants of MiniMax H3 post-trained by fal for prompt adherence, aesthetics and inference speed. H3 Max also takes reference-to-video.
The models are hosted in the US, and Microsoft says Foundry is the first provider to offer a US-hosted Kling endpoint.
Prices
Microsoft lists preview prices per second of generated video:
| Model | Resolution | US$ per second |
|---|---|---|
| MiniMax H3 Max | 480p / 768p / 1080p | 0.05 / 0.08 / 0.16 |
| MiniMax H3 Max Turbo | 480p / 768p / 1080p | 0.025 / 0.04 / 0.08 |
| LTX-2.5 Fast | 720p / 1080p / 1440p / 4K | 0.09 / 0.13 / 0.19 / 0.30 |
| LTX-2.5 Pro | 720p / 1080p | 0.12 / 0.17 |
| Kling v3 Pro | audio off / audio on / audio + voice control | 0.1232 / 0.1848 / 0.2156 |
Microsoft says the prices will move to its Azure pricing page in the coming weeks. The post does not give clip lengths.
Why fal is involved
fal describes itself as a full stack for generative media, with APIs for more than 1,500 models across image, video, audio, 3D and “generative worlds”, plus inference, training and infrastructure. Microsoft’s argument is that video work is rarely served by one model: one brief may need several interpretations, another a product image turned into a short motion test, another a pipeline from text to a frame to a scene. The aim is to let developers choose a model for each step and connect the steps in one application.
Limits
- It is a preview, so expect changes to model versions, quotas and pricing.
- Microsoft’s post is a partnership announcement. It gives no benchmarks, and the descriptions of each model are the vendors’ own.
- Kling is listed as a “US Data Zone” deployment, while the LTX and MiniMax models are “Global Standard”. Teams with data-residency rules should check each model’s terms.
Our take
Distribution is becoming part of the story. When several competing video models sit behind one enterprise account, studios can swap models without rebuilding their pipeline, which suits interactive film and game cutscenes where the best model changes month to month. It does not make the models interactive; for that, see generative interactive film. It does lower the cost of trying them.