Running ComfyUI workflows on Amazon SageMaker AI processing jobs
TL;DR
AWS published a CDK blueprint for running ComfyUI as a batch pipeline on SageMaker AI processing jobs: Lambda triggers the job, ECR serves the container, and outputs stream to S3. The sample uses Z-Image Turbo in a ComfyUI workflow, six ml.g5.xlarge GPU instances, 125 GB volumes, VPC private subnets, KMS encryption, and CloudWatch logs. The job reads prompts from a file, injects prompt and seed into a workflow template, sends requests to ComfyUI, and polls the queue every 15 seconds until done.
Nauti's Take
This is not a magic creative autopilot. It is an infrastructure recipe for teams already using ComfyUI seriously.
The strong part: jobs shut down automatically, outputs stream to S3, logs are available, and custom workflows can be swapped in. The catch: anyone who just wants to generate a few images may create an expensive cloud stack of CDK, Docker, quotas, and GPU instances.
It makes sense when variant production, compliance, and repeatable runs matter more than quick experiments.
Briefingshow
ComfyUI has become a practical frontend for image and creative workflows, but local setups break down when teams need hundreds of variants. AWS frames it as a controlled batch job with logs, network isolation, and automatic GPU shutdown. The important shift is not the demo itself, but the move from a designer tool to a repeatable production job.
Sources
