Fal models on Microsoft Foundry (preview)

By using the integration with fal, Microsoft Foundry customers can use five video generation models in public preview:

  • Generate video from text or images with Kling v3 Pro, LTX v2.5 Fast and Pro, and MiniMax H3 Max and Max Turbo.
  • Use reference-to-video with MiniMax H3 Max.
  • Choose a deployment offering: US Data Zone for Kling v3 Pro, or Global Standard pay-as-you-go (PayGo) for the LTX and MiniMax models.

Compare the models and their modalities in Available catalog models.

Important

Microsoft is providing you with an optional capability to access fal services on Microsoft Foundry. Fal services are operated outside of Microsoft facilities. By selecting to deploy this model, you agree on behalf of your organization that you are bound by and must use the fal services only in accordance with fal Terms of Service. In doing so, you are electing to share your organization's data with fal and egress data from Microsoft. Your Microsoft customer agreements, including Product Terms, DPA, and certifications do not apply to your use of Fal services on Microsoft Foundry.

Prerequisites

  • An Azure subscription. If you don't have one, create a free account.

  • A Foundry resource with a Foundry project.

  • An Azure identity with the Subscription Owner or Subscription Contributor role to enable the feature.

  • To deploy models, you need the Foundry Owner role on the Foundry project. For more information, see Azure built-in roles.

    Important

    The Foundry RBAC roles were recently renamed. Foundry User, Foundry Owner, Foundry Account Owner, and Foundry Project Manager were previously named Azure AI User, Azure AI Owner, Azure AI Account Owner, and Azure AI Project Manager. You might still see the previous names in some places while the rename rolls out. The role IDs and core permissions are unchanged by the rename.

Offering availability

Kling v3 Pro supports US Data Zone. LTX v2.5 Fast, LTX v2.5 Pro, MiniMax H3 Max, and MiniMax H3 Max Turbo support Global Standard PayGo. For general deployment-type information, see Deployment types.

Enable fal on Foundry

Review the acknowledgment at the top of this article before enabling fal. To enable fal on Microsoft Foundry, an administrator must enable the service within your Azure subscription. The Azure Preview features portal enables customers to turn on the service by using the following steps.

  1. Sign in to the Azure portal.

  2. In the search box, enter subscriptions and select Subscriptions.

  3. Select the link for your subscription's name.

  4. From the left menu, under Settings, select Preview features.

  5. Search for and select the Fal.EnableDeploy preview feature.

  6. Review the terms provided in the Description and the data privacy section in this documentation.

  7. If you don't agree to the terms, select Close and don't continue. Otherwise, select Register.

  8. Select OK. The Preview features screen refreshes and the preview feature's State is displayed. It might take up to 30 minutes for the feature to enable for your subscription.

    Tip

    To verify registration, refresh the Preview features page and confirm the State column shows Registered for the fal on Foundry feature.

Deploy fal models from the Foundry portal

After the feature is enabled, you can deploy fal models from the Foundry model catalog. Complete these steps to get a live endpoint for chat completions. Choose a model from Available catalog models. For the general portal workflow, see Deploy Foundry Models in the portal.

  1. From the Foundry portal homepage, select Discover in the upper-right navigation.

  2. In the left pane, select Models to open the model catalog.

  3. Select the fal model you want to deploy and review its model card.

  4. On the model page, select Deploy. For more information on deployment options, see Deploy Foundry Models in the portal.

  5. In the deployment window, configure the following settings:

    • Deployment name: Keep the default name or enter a custom name to identify the deployment.
    • Token Plan: Select Pay-per-token -> Data Zone Standard or Global Standard, depending on the model.
    • Model version: Select the model version for the deployment.

  6. Select Deploy. The deployment process can take up to 30 minutes.

  7. After deployment completes, use the provided endpoint and key to send inference requests to the model. To quickly test the deployment, use the Playground in your Foundry project.

    Tip

    To verify the deployment, navigate to your project's Deployments page and confirm the deployment Status shows Succeeded.

Use the video generation APIs

Fal on Microsoft Foundry models supports the following API routes:

  • Kling V3 Pro (text-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/kling/v3/pro/text-to-video
  • Kling V3 Pro (image-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/kling/v3/pro/image-to-video
  • LTX 2.5 Fast (text-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/lightricks/ltx-2.5/text-to-video/fast
  • LTX 2.5 Fast (image-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/lightricks/ltx-2.5/image-to-video/fast
  • LTX 2.5 Pro (text-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/lightricks/ltx-2.5/text-to-video/pro
  • LTX 2.5 Pro (image-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/lightricks/ltx-2.5/image-to-video/pro
  • MiniMax H3 Max (text-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/minimax/h3-max/text-to-video
  • MiniMax H3 Max (image-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/minimax/h3-max/image-to-video
  • MiniMax H3 Max (reference-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/minimax/h3-max/reference-to-video
  • MiniMax H3 Max Turbo (text-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/minimax/h3-max-turbo/text-to-video
  • MiniMax H3 Max Turbo (image-to-video): Uses https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/minimax/h3-max-turbo/image-to-video

For more information on supported parameters and the API schema, please review the fal models documentation.

Submit and retrieve a video (text-to-video)

The below example bash script utilizes MiniMax H3 Max with the text-to-video modality. Ensure you have updated the FOUNDRY_RESOURCE_NAME and the DEPLOYMENT_NAME to reflect your Foundry project endpoint name and model deployment name.

!/bin/bash 
 
BASE_URL="https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/v1/minimax/h3-max" 
get_bearer() { 
  az account get-access-token \ 
    --scope 'https://cognitiveservices.azure.com/.default' \ 
    --query accessToken --output tsv 
} 
 
# Create the video, make sure the model matches the deployment name 
REQUEST_ID="$(curl --fail-with-body --verbose --show-error \ 
  "${BASE_URL}/text-to-video" \ 
  -H 'Content-Type: application/json' \ 
  -H "Authorization: Bearer $(get_bearer)" \ 
  --data '{ 
    "model": "<DEPLOYMENT_NAME>", 
    "prompt": "A cat climbing a tree.", 
    "duration": 5, 
    "resolution": "480P", 
    "prompt_expansion_mode": "disabled" 
  }' | jq -r '.request_id')" 
 
# Poll until status completed 
while true; do 
  STATUS=$(curl --fail-with-body --silent --show-error \ 
    "${BASE_URL}/requests/${REQUEST_ID}/status" \ 
    -H "Authorization: Bearer $(get_bearer)" \ 
    | jq -r '.status') 
  echo "Current status: $STATUS" 
  if [ "$STATUS" = "COMPLETED" ]; then 
    break 
  fi 
  sleep 5 
done 
echo "Video creation completed. Request ID: $REQUEST_ID" 
 
# Get the video URL 
VIDEO_URL="$(curl --fail-with-body --verbose --show-error \ 
  "${BASE_URL}/requests/${REQUEST_ID}" \ 
  -H "Authorization: Bearer $(get_bearer)" \ 
  | jq -r '.video.url')" 
echo "Video URL: $VIDEO_URL" 
 
# Download the video 
curl --fail-with-body --show-error --location \ 
  "$VIDEO_URL" \ 
  -H "Authorization: Bearer $(get_bearer)" \ 
  --output video.mp4 
 
# Play the downloaded video using VLC 
vlc video.mp4

Submit and retrieve a video (image-to-video)

The below example bash script utilizes LTX 2.5 Pro with the image-to-video modality. Ensure you have updated the FOUNDRY_RESOURCE_NAME, IMAGE_URL and the DEPLOYMENT_NAME to reflect your Foundry project endpoint name, the image you wish to use, and model deployment name.

#!/bin/bash
set -euo pipefail

BASE_URL="https://{FOUNDRY_RESOURCE_NAME}.services.ai.azure.com/providers/falai/v1/lightricks/ltx-2.5"
IMAGE_URL="{IMAGE_URL}"
DEPLOYMENT_NAME="{DEPLOYMENT_NAME}"

get_bearer() {
  az account get-access-token \
    --scope 'https://cognitiveservices.azure.com/.default' \
    --query accessToken --output tsv
}

REQUEST_ID="$(curl --fail-with-body --verbose --show-error \
  "${BASE_URL}/image-to-video/fast" \
  -H 'Content-Type: application/json' \
  -H "Authorization: Bearer $(get_bearer)" \
  --data '{
    "model": "'$DEPLOYMENT_NAME'",
    "prompt": "Please continue the video from the input image.",
    "image_url": "'$IMAGE_URL'",
    "duration": 6,
    "resolution": "720p",
    "aspect_ratio": "16:9"
  }' | jq -r '.request_id')"

echo "Request ID: $REQUEST_ID"

while true; do
  STATUS="$(curl --fail-with-body --silent --show-error --max-time 60 \
    "${BASE_URL}/requests/${REQUEST_ID}/status" \
    -H "Authorization: Bearer $(get_bearer)" \
    | jq -r '.status')"
  echo "Current status: $STATUS"
  if [ "$STATUS" = "COMPLETED" ]; then
    break
  fi
  sleep 5
done
echo "Video creation completed. Request ID: $REQUEST_ID"

VIDEO_URL="$(curl --fail-with-body --silent --show-error --max-time 60 \
  "${BASE_URL}/requests/${REQUEST_ID}" \
  -H "Authorization: Bearer $(get_bearer)" \
  | jq -er '.video.url | strings | select(length > 0)')"
echo "Video URL: $VIDEO_URL"

curl --fail-with-body --show-error --location --max-time 120 \
  "$VIDEO_URL" \
  -H "Authorization: Bearer $(get_bearer)" \
  --output "video.mp4"
echo "Video saved to: video.mp4"

Quotas and rate limits

The following table lists rate limits in requests per minute (RPM) for the free tier and quota tiers 1 through 7.

Model Free tier Tier 1 Tier 2 Tier 3 Tier 4 Tier 5 Tier 6 Tier 7
Kling v3 Pro 0 2 2 2 2 40 40 40
LTX v2.5 Fast 0 2 2 2 2 40 40 40
LTX v2.5 Pro 0 2 2 2 2 40 40 40
MiniMax H3 Max 0 2 2 2 2 40 40 40
MiniMax H3 Max Turbo 0 2 2 2 2 40 40 40

Available catalog models

These video generation models support text-to-video and image-to-video. MiniMax H3 Max also supports reference-to-video.

Model Supported modalities Deployment offering
Kling v3 Pro Text-to-video, image-to-video US Data Zone
LTX v2.5 Fast Text-to-video, image-to-video Global Standard PayGo
LTX v2.5 Pro Text-to-video, image-to-video Global Standard PayGo
MiniMax H3 Max Text-to-video, image-to-video, reference-to-video Global Standard PayGo
MiniMax H3 Max Turbo Text-to-video, image-to-video Global Standard PayGo
  • Text-to-video: Generate video from a text prompt.
  • Image-to-video: Generate video using an input image and a prompt.
  • Reference-to-video: Use reference images, videos, or audio with MiniMax H3 Max.

Generation settings

The following settings describe fal's public model APIs for text-to-video and image-to-video. MiniMax H3 Max's reference-to-video API uses the same resolution and duration ranges.

Model Resolutions Requested duration (seconds) Frame rates (FPS)
Kling v3 Pro Not specified in the API schema. 3 to 15, in one-second increments. Not specified in the API schema.
LTX v2.5 Fast 720p, 1080p, 1440p, 2160p 6, 8, 10, 12, 14, 16, 18, 20, or auto, subject to the limits below. 24, 25, 48, 50
LTX v2.5 Pro 720p, 1080p 6, 8, 10, or auto 24, 25, 50
MiniMax H3 Max 480P, 768P, 1080P 0.92 to 15 N/A
MiniMax H3 Max Turbo 480P, 768P, 1080P 0.92 to 15 N/A

Key model differences and constraints:

  • Kling v3 Pro supports native audio generation and multi-shot prompts. Text-to-video supports 16:9, 9:16, and 1:1 aspect ratios. Image-to-video accepts a start image and an optional end image.
  • LTX v2.5 Fast is optimized for rapid iteration. At 720p or 1080p, 24 or 25 FPS supports up to 20 seconds; 48 or 50 FPS supports up to 10 seconds. At 1440p or 2160p, all listed frame rates support up to 10 seconds.
  • LTX v2.5 Pro is optimized for output quality, with a maximum duration of 10 seconds. Both LTX variants support synchronized video and audio, and an optional end image for image-to-video.
  • MiniMax H3 Max supports reference-to-video with up to 12 reference files across images, videos, and audio. Reference video and audio clips must each be 2 to 15 seconds. The combined video duration and combined audio duration each have a 15-second limit.
  • MiniMax H3 Max and Max Turbo default to a five-second duration and 768P resolution. 1080P uses latent refinement from a native 768P source. Generated videos can run approximately 0.7 seconds longer than requested.

Data privacy

Review the acknowledgement and fal Terms of Service against your organization's data-handling and compliance requirements before sending requests.

Transparency note

Assess each model's suitability for your workload before using its outputs. For general evaluation guidance, see Risk and safety evaluations.

Frequently asked questions

The following questions cover launch offerings and setup.

Which deployment offerings are available?

Kling v3 Pro launches with US Data Zone. The four LTX and MiniMax models launch with Global Standard PayGo.

Which models support reference-to-video?

MiniMax H3 Max supports reference-to-video. The other four launch models support text-to-video and image-to-video.

Can I use existing fal deployments or credentials?

No; you can't use your existing fal deployments on the fal platform with fal on Microsoft Foundry. While the models might be the same, you must create a new Foundry deployment to use the models with fal on Microsoft Foundry.

Related content

  • Last updated on