Running ComfyUI workflows on Amazon SageMaker AI processing jobs

This article demonstrates how to deploy ComfyUI workflows on Amazon SageMaker AI processing jobs to automate large-scale image generation for enterprises using GPU-accelerated infrastructure.

• ComfyUI is a node-based visual workflow builder for generative AI that enables reproducible image, audio, and video pipelines without coding
• Solution uses Z-Image Turbo, a 6B-parameter diffusion transformer model, for fast, high-quality text-to-image generation
• Architecture comprises three AWS CDK stacks: DataStack (S3 storage), SecurityStack (VPC with encryption), and ComfyUISmStack (Lambda trigger and SageMaker processing)
• Processing jobs run on GPU-accelerated ml.g5.xlarge instances in private VPC subnets with real-time image streaming to S3
• Step-by-step deployment guide includes prerequisites, environment setup, AWS CDK bootstrap, and Lambda function invocation
• Use cases include A/B testing ad creatives, locale-specific packaging design, and interactive video storytelling at scale

By combining ComfyUI with SageMaker AI's managed GPU infrastructure and pay-per-second billing, enterprises can automate content generation at scale with built-in security and cost efficiency.