Deployment FAQ
Direct answers about workflow compatibility, startup, billing, persistence, and support.
What is ModelPilot for?
ModelPilot helps you inspect, download, configure, and deploy open-source models and ComfyUI workflows without operating the GPU infrastructure yourself. Start with a workflow guide that names its model files, node requirements, and remaining manual steps.
Do I need technical experience?
You do not need to manage Docker or rent a GPU directly. You should still understand the workflow you are running, review its model and custom-node requirements, read deployment warnings, and stop the GPU when finished. Arbitrary custom workflows may require ComfyUI troubleshooting.
Which workflows are supported?
Browse the workflow library and read each guide’s evidence and known limitations. The workflow checker can analyze other ComfyUI frontend or API JSON, but a successful analysis does not guarantee that every custom-node version, gated model, resolution, or batch size will run.
What happens to workflow JSON submitted to the checker?
The web checker sends the JSON securely to ModelPilot for server-side analysis and does not store it. Referenced image, video, output, and model-weight files are not uploaded. Remove passwords, tokens, private URLs, and other secrets from node fields first.
How long does a deployment take?
Startup takes minutes rather than seconds. Cached image or text configurations can be ready in several minutes; large video models and custom workflows commonly take 15–35 minutes or longer. GPU availability, image caching, model downloads, and custom nodes all affect the result. Dashboard status is authoritative; estimates are not an uptime or startup SLA.
When does dedicated-GPU billing start and stop?
Dedicated compute is charged proportionally for billable running GPU time, including model downloads and startup while the GPU is running. There is no rounding to a 10-minute billing block. A deployment never placed on a GPU is excluded; failure after allocation can still incur charges. Stopping ends compute billing, but retained storage may still cost money.
Will a deployment stop when it is idle?
No. You must stop an idle dedicated deployment from the dashboard. Low-balance protection stops deployments when prepaid credit runs out, but it is a financial safety net rather than an idle timeout.
How do prepaid credits work?
Credit top-ups start at $5. Dedicated deployments draw down credit according to actual running time; serverless models are billed per successful request. Current rates and available purchase amounts are shown on the pricing page and in the billing screen before checkout.
Are all models available through the serverless API?
No. Serverless has a deliberately narrow catalog. Check the current serverless catalog before integrating. Wan video is currently dedicated-only; its serverless endpoint is not live.
Do files persist after stopping?
Stopping can retain the deployment disk and its storage charges; it is not the same as deleting. Longer-term persistence depends on the configured storage and which directories are mounted. Download needed outputs before deleting a deployment or its storage.
What should I do if deployment fails?
- Open the deployment logs from the dashboard.
- Find the first model-download, custom-node import, or GPU-memory error rather than the later cascade of failures.
- Correct the workflow requirement or GPU choice before retrying.
- Use the troubleshooting guide.
How do I contact support?
Email support@modelpilot.ai with the deployment ID, the first relevant log error, the workflow page or model selected, and the approximate failure time. Do not email passwords, API keys, private model tokens, or complete workflows containing secrets.