ComfyUI just made one of its biggest cloud workflow upgrades.
You can now package your own custom ComfyUI environment, including custom nodes, models, LoRAs, and Python dependencies, and deploy it to Comfy Cloud.
The result? Your artists can use the same setup in their browsers, while developers run workflows through an API.
No more asking everyone on the team to install exactly the same collection of custom nodes.
WHAT'S NEW IN COMFY CLOUD?
Custom Builds bring your own ComfyUI setup into a managed cloud environment.
You can:
• Package supported public or private custom nodes.
• Bring your models, LoRAs, and their dependencies.
• Run different workflows using the same deployed environment.
• Adjust prompts, parameters, nodes, and connections in the browser.
• Add a new LoRA or node pack, deploy an updated Build, and roll it out to the workspace.
• Run workflows programmatically through the Developer Platform API.
• Scale GPU workers as demand increases and let them scale back down afterward.
The setup can be created from an existing workflow or Comfy Desktop snapshot through the Developer Platform. Complex dependencies and unsupported nodes may still need manual troubleshooting.
ONE BUILD, MULTIPLE WORKFLOWS
A Build is the environment: the available nodes, models, and dependencies.
A workflow is what you actually do with those tools.
You can edit and create multiple workflows inside the same Build without rebuilding it, as long as they don't need additional nodes or models.
When the underlying environment changes, you deploy a new version. Existing jobs finish on the version they started with, and you can roll back if needed.
GPU OPTIONS AND PRICING
Comfy lists these custom Build GPU rates:
• RTX PRO 6000 MIG, 48 GB: $1.75/hour.
• RTX PRO 6000, 96 GB: $3.49/hour.
• H100 SXM, 80 GB: $6.23/hour.
• H200 SXM, 141 GB: $7.71/hour.
• B200, 180 GB: $11.23/hour.
Compute is billed by the GPU second, not by the hour-long block.
Opening the editor, connecting nodes, or adjusting a workflow doesn't start GPU billing on its own.
But there's a catch: you're billed whenever a worker is running, including the keep-alive window after a job and any workers deliberately kept warm.
The default keep-alive period is three minutes. On a B200, three idle minutes cost about $0.56.
Persistent model storage is separate: $0.20 per GB per month. Keeping 20 GB of models costs $4 per month before compute.
WHICH PLAN DO YOU NEED?
Custom Builds are available on paid Comfy Cloud plans, starting with Standard.
• Standard and Creator: up to 2 concurrent jobs per deployment.
• Pro: up to 10.
• Team: up to 20, plus the ability to invite workspace members.
Team members can use the same deployed environment, but access to a Build doesn't automatically share everyone's individual workflows or outputs.
WHY THIS MATTERS
A custom ComfyUI installation is rarely just a collection of workflow JSON files.
It's an entire software environment: specific node versions, dependencies, checkpoints, LoRAs, and GPU requirements.
Making that environment reproducible and accessible through both a visual editor and an API is a meaningful improvement for production teams.
It could help with AI-assisted filmmaking, image generation, product photography, animation pipelines, and batch content production.
You still have to check node compatibility, GPU memory, cost, and the licensing of your models.
And a cloud deployment won't be automatically cheaper than a local GPU, especially if you leave workers warm between jobs.
But the direction is clear: ComfyUI is becoming more than a local node-based creative tool. It's also a deployable workflow environment for teams and applications.
SOURCES
Official announcement:
https://blog.comfy.org/p/your-custom-comfyui-is-now-in-comfy
Comfy Cloud:
https://comfy.org/cloud
Custom Builds documentation:
https://docs.comfy.org/development/deploy/cloud-custom-builds
GPU pricing:
https://comfy.org/pricing/
Would you move your custom ComfyUI setup to the cloud, or would you rather keep everything local?
#ComfyUI #ComfyCloud #AIWorkflows
YouTube / ABVX
