Skip to main content

Documentation

Documentation/Custom Workflows

Custom Workflows

Deploy your own ComfyUI workflows with automatic custom node installation and model downloading. Perfect for complex pipelines with ControlNet, LoRA, IPAdapter, and more.

What is Workflow Deployment?

Instead of selecting a pre-configured model, you can upload your own ComfyUI workflow JSON file. ModelPilot automatically:

  • Detects required custom nodes and installs them
  • Identifies models referenced in the workflow
  • Downloads models from Hugging Face or CivitAI
  • Configures the appropriate GPU based on requirements
  • Pre-loads your workflow so it opens ready to use

How to Deploy a Workflow

  1. Export your workflow from ComfyUI

    In ComfyUI, go to Menu → Save (or press Ctrl+S) to download the workflow JSON

  2. Start a new deployment

    Go to the deployment wizard and select the Workflow tab

  3. Upload your workflow file

    Drag and drop or click to select your .json file

  4. Review the analysis

    ModelPilot shows detected custom nodes, models, and recommended GPU

  5. Deploy

    Click deploy and wait for setup (custom workflows may take 10-20 minutes)

Supported Features

Custom Nodes

ComfyUI Manager nodes, ControlNet, IPAdapter, AnimateDiff, KSampler variants, and more. Nodes are auto-detected from your workflow.

Model Downloads

Checkpoints, LoRAs, VAEs, ControlNet models, and upscalers from Hugging Face. Embed download URLs in your workflow for automatic setup.

Base Architectures

Flux, SDXL, SD 1.5, Wan Video, HunyuanVideo, and more. Auto-detected from your workflow nodes.

Best Practices

Embedding Model URLs

For automatic model downloading, include download URLs in your workflow. In ComfyUI:

  1. Right-click on a model loader node
  2. Select Properties Panel or Edit
  3. Add the Hugging Face URL for the model
  4. Save and export the workflow

Example URL Format

https://huggingface.co/black-forest-labs/FLUX.1-dev/resolve/main/flux1-dev.safetensors

Testing Before Deployment

  • Test your workflow locally first to ensure it works
  • Make sure all custom nodes are from ComfyUI Manager
  • Verify model paths use relative names (not absolute paths)
  • Keep workflows simple for faster deployment

GPU Selection

ModelPilot recommends a GPU based on your workflow, but you can override:

  • L4 (24GB): SD 1.5, simple SDXL, Flux Schnell
  • A6000 (48GB): Flux Dev, complex SDXL with ControlNet, Z Image
  • A100 (80GB): Video models, very large batches

Troubleshooting

Workflow not loading?

Check that all custom nodes were detected. Some private or very new nodes may not be recognized. You can manually add custom node Git URLs in the advanced settings.

Models not found?

Ensure model names in your workflow match standard naming conventions. For custom models, embed the download URL in the workflow properties.

Deployment taking too long?

Complex workflows with many custom nodes and large models can take 15-25 minutes. Check the deployment logs in your dashboard for progress.

Out of memory?

Choose a larger GPU or simplify your workflow. Video generation and large batches require more VRAM than single image generation.

Ready to deploy your workflow?

Start by trying a Quick Deploy to understand the process, then upload your custom workflow.

Deploy Custom Workflow