Generate images and video with vLLM-Omni on SageMaker AI – Part 2

This article demonstrates how to generate images and animate them into videos using vLLM-Omni on Amazon SageMaker AI, continuing a series on specialized AWS Deep Learning Containers.

• Deploy FLUX.2-klein-4B for real-time image generation via a SageMaker real-time endpoint
• Use Wan2.1-VACE-1.3B for video generation via a SageMaker Asynchronous Inference endpoint
• AWS vLLM-Omni DLC packages vLLM-Omni releases with routing middleware for SageMaker AI
• Real-time endpoint returns base64-encoded PNG directly; asynchronous endpoint writes MP4 to Amazon S3
• Workflow converts generated image to JPEG data URL and passes it to video model as image-conditioned input
• Sample includes command-line workflow and Streamlit application for end-to-end generation
• Image endpoint deployed on ml.g6.xlarge; video endpoint on ml.g6e.xlarge instance types

The solution demonstrates how a shared serving runtime supports different generative media models with appropriate inference patterns for each workload.