ComfyUI video generation at scale: how to build a 30-50 videos/day pipeline
Admirable-Card-2965 · reddit · 2026-09-16
A ComfyUI newcomer is asking how to scale video generation: with a Minimax H3 workflow on a cloud GPU, a 6-8 second clip takes 20-30 minutes, and he often needs 2-5 generations to get one usable video, so sequential generation can't support 30-50+ videos per day.
His open questions:
- Run multiple GPU instances/pods in parallel?
- Build a job queue that distributes tasks across GPU workers from one workflow?
- Tools/APIs to spin GPU instances up and down based on queue depth? RunPod Serverless vs. persistent pods?
- How to handle the 2-5x retry overhead economically?
His ideal setup: submit 100 prompts/images, have them auto-distributed across available GPUs, generate in parallel, then shut GPUs down when the queue drains — prioritizing cost efficiency over latency.
More from Infra
- Dev slams agent sandbox pricing as 20x+ the cost of a $6/month always-on VPS — Aryvyo · 2026-09-16
- Signal65 benchmarks NVIDIA Vera CPU at 1.64x per core vs x86 on agentic workloads — ryanshrout · 2026-09-16
- McKinsey: Sovereign AI TAM to hit $500-600B by 2030, 30-40% of AI demand — Beth_Kindig · 2026-09-16
- Intel CEO Lip Bu Tan tells AI Infra Summit: embrace AI, don't fear it — karlfreund · 2026-09-16
- Brad Gerstner: AI buildout hinges on revenue keeping its steep growth curve — markjeffrey · 2026-09-16
- Reported: NVIDIA hardware delivers 40% more throughput at the same power — karlfreund · 2026-09-16