InjazAI

AWS enabled running two generative media models — FLUX.2-klein for images and Wan2.1-VACE for video — from a single vLLM-Omni container on Amazon SageMaker AI, using real-time inference for images and asynchronous inference for video with output saved to Amazon S3.

1 min read

vLLM-Omni on SageMaker AI: Image and Video Generation from One Container

What was published

FAQ

What is vLLM-Omni on SageMaker AI?

It is an AWS Deep Learning Container built on vLLM-Omni that lets teams run generative media models (images and video) on Amazon SageMaker AI without building a custom inference environment per model.

Why use real-time inference for images and asynchronous inference for video?

FLUX.2-klein image generation completes in seconds, so real-time inference fits. Wan2.1-VACE video generation takes longer, so it runs asynchronously, with the output stored in Amazon S3 and retrieved later.

Does this pattern help MENA AI teams?

Yes. It cuts the number of containers and endpoints needed to run a full generative media pipeline, simplifying operations and cost control for media, marketing and government teams building visual content on AWS.

Is this a new model announcement?

No. It is the second part of a hands-on series explaining how to deploy existing models (FLUX.2-klein and Wan2.1-VACE) from one vLLM-Omni container on SageMaker AI.

Source: AWS Machine Learning

AI-assisted summary of the source above. High-confidence stories publish automatically; others wait for review. Facts that are not in the source are not added.