Quick Deploy Guide

Quick Deploy starts a preconfigured model or verified workflow on a dedicated cloud GPU. It removes most infrastructure setup, but the first start still takes minutes while the GPU, container, models, and custom nodes become ready.

Before you deploy

• Browse a verified workflow when possible. These have explicit files, requirements, and known limitations.
• Use the free workflow checker before uploading an unfamiliar ComfyUI workflow.
• Deployment requires an account and prepaid credits. Top-ups start at $5.
• Dedicated deployments do not stop merely because they are idle. Stop the deployment from your dashboard when you finish.

Step by step

1. Choose a starting point. Select a verified workflow, a catalog model, or upload your own ComfyUI JSON.
2. Review compatibility. Check detected model filenames, custom nodes, warnings, and the recommended GPU. A green check means the required items were recognized; it is not a guarantee that every graph will execute successfully.
3. Review cost and storage. Confirm the hourly GPU rate, minimum balance, and whether you need optional persistent storage.
4. Start the deployment. The dashboard shows setup and startup progress. Billing begins when the deployment reaches Running, not while it is being prepared.
5. Open the interface. Use the dashboard link after the health check reports the interface is ready.
6. Stop when finished. Stopping ends compute billing. If persistence was not enabled, download the outputs you need before stopping or deleting the deployment.

What ModelPilot configures

• A recommended GPU for recognized requirements
• The container and supported model runtime
• The applicable ComfyUI, model, or API interface
• Optional persistent storage when selected
• Deployment status, logs, and stop controls

Expected startup time

A cached image model can become ready in several minutes. Large video models and custom workflows can take 15–35 minutes or longer when model files must be downloaded. GPU availability, model size, gated downloads, and custom-node installation can all change the time. Use the live dashboard status instead of treating an estimate as an SLA.

When the self-serve path is not a good fit

• Private or unlisted custom-node repositories
• Gated model downloads without usable credentials
• Multi-GPU or unusual networking requirements
• A workflow that has not first worked in ComfyUI
• Workloads that require a formal uptime SLA