​
Video Generation Guide: Using ComfyUI on Vast.ai

This guide will walk you through setting up and using ComfyUI for video generation on Vast.ai. ComfyUI provides a powerful node-based interface for creating advanced stable diffusion pipelines, making it ideal for video generation workflows.

​
Prerequisites

​
Setting Up Your Instance

​
1. Select the Right Template

Navigate to the Templates tab to view available templates. For video generation, we recommend searching for “ComfyUI” among the recommended templates. The ComfyUI template provides a powerful and modular stable diffusion GUI for designing and executing advanced pipelines using a graph/nodes/flowchart based interface. Template Features:
  • Access through both Jupyter and SSH
  • Instance Portal
  • Token-based authentication enabled by default
  • Built-in provisioning script for models and custom nodes

​
2. Edit your Template Configuration

Add/update these environment variables as needed:
Bash
# Core Settings
COMFYUI_ARGS="--disable-auto-launch --port 18188 --enable-cors-header"         # ComfyUI launch arguments

# Authentication
WEB_ENABLE_HTTPS=false   # Enable/disable direct HTTPS
WEB_ENABLE_AUTH=true    # Enable/disable authentication

# Access Tokens
CF_TUNNEL_TOKEN=""      # Cloudflare Zero Trust token
CIVITAI_TOKEN=""        # Access gated Civitai models
HF_TOKEN=""            # Access gated HuggingFace models

# Custom Setup
PROVISIONING_SCRIPT=""  # URL to custom provisioning script
Provisioning Script:
  • Default script includes popular image models and custom nodes
  • Fully customizable - Create your own script for a custom instance
  • Must be Bash-compatible and start with #!/bin/bash
  • Upload modified script to a GitHub Gist or respository and update the PROVISIONING_SCRIPT variable to point to the raw file
Important: Never save your template as public if you’ve included tokens or other secrets in the Docker Options field.
Select your template from ‘My Templates’ after making any desired edits to it.

​
3. Create Your Instance

  1. In the Search interface, look for machines that have sufficient VRAM to handle your chosen video model. ⚠All models are different so check the model requirements carefully.
  2. Click RENT to create an instance on the machine with the GPU of your choice

​
3. Connect to Your Instance

  1. Go to Instances tab to see your instance loading
  2. When the blue button says “OPEN”, click this button to access the Instance Portal which will provide access to ComfyUI and other useful applications.
  3. Click the direct link or cloudflare quick tunnel link to access ComfyUI. Here’s a beginner’s guide to using ComfyUI.

​
4. Select a Video Workflow

ComfyUI has a workflow browser, so for a quick start you can choose on of their templates
ComfyUI template workflows

ComfyUI Workflows

We’ll pick the LTX Video workflow for this guide. Simply click it to proceed.

​
5. Download Missing Files

Your new instance will not yet have the required models, but fortunately ComfyUI will alert us to this and offer the models for download. ComfyUI template workflows Unfortunately, the interface does not know it is running in the cloud so clicking the download buttons will download the models to your local machine. To work around this you can either:
  • Download the models to your computer and then upload them to the instance
  • SSH to the instance and use curl or wget to directly download the models to their correct locations
To complete this guide we will use SSH
Bash
# Download the models
wget --content-disposition -P /workspace/ComfyUI/models/checkpoints/ "https://huggingface.co/Lightricks/LTX-Video/resolve/main/ltx-video-2b-v0.9.safetensors"
wget --content-disposition -P /workspace/ComfyUI/models/clip/ "https://huggingface.co/comfyanonymous/flux_text_encoders/resolve/main/t5xxl_fp16.safetensors"

# Download the workflow source image
wget --content-disposition -P /workspace/ComfyUI/input/ "https://comfyanonymous.github.io/ComfyUI_examples/ltxv/island.jpg"
The above commands will download the required models into the instance. When the downloads have completed you can refresh the browser window to clear the missing models error.

​
6. Run the Workflow

Finally, click the Run button to process the workflow. Feel free to modify the prompts and experiement!

​
Pre-Configured Templates

We have some pre-configured ComfyUI templates - And one for this guide. Check them out here

​
Resources and Further Reading

  1. ComfyUI Official Repository
  2. Vast.ai Documentation
  3. Comfy Workflows
  4. Vast.ai support chat on website
Remember to always check VRAM usage and adjust parameters accordingly. Start with smaller frames and resolutions, then scale up as you become more comfortable with the workflow.