Running ComfyUI workflows on Amazon SageMaker AI processing jobs
TL;DR
AWS shows how to run ComfyUI workflows as Amazon SageMaker AI Processing Jobs to generate hundreds of images in one batch. The reference architecture uses AWS CDK, Lambda as the trigger, ECR for the container, S3 for outputs, CloudWatch for logs, plus VPC and KMS for isolation and encryption. The demo runs a custom ComfyUI container with Z-Image Turbo on six ml.g5.xlarge GPU instances with a 125 GB volume.
Nauti's Take
AWS frames this as an enterprise creative pipeline, and some of that is obvious PR packaging. Still, the pattern is useful: ComfyUI remains the visual workflow builder, while SageMaker handles controlled GPU execution.
For small teams, this is probably too much infrastructure. For agencies, ecommerce teams, or media companies with many variants, fixed prompts, and compliance pressure, this batch setup can make sense.
Briefingshow
For teams already using ComfyUI locally, this is a concrete path from desktop workflow to repeatable batch production. The important part is not just more GPU power; operations matter: containers, quotas, logs, S3 outputs, network control, and automatic shutdown decide whether image generation becomes reliable production or an expensive cloud experiment.
Sources
