Sign In
home
models
images
videos
posts
articles
bounties
tools
challenges
events
shop

Guide to RUNPOD Template - LatentDream's Hunyuan 💥 AllInOne

19
DIhan's Avatar
DIhan
Mar 1, 2025
generation guide
Guide to RUNPOD Template - LatentDream's Hunyuan 💥 AllInOne

# 🎮 ComfyUI All-in-One RunPod Template

Big shoutout the @latetdream 's awesome workflow series. Although his method works for basic GPU's you can really get incredible results with a monter GPUs at Runpod. $0.44 an hour on a 48GB is fun.

TEMPLATES
💥3.0-CUDA2.5 (More GPU's avaliable AllinOne Ultra 1.4 Support - Recommended)
💥3.0-CUDA2.6 (Not as many GPU avaliable and slightly better performance)

Welcome to the most awesome ComfyUI setup you'll ever encounter! This template comes packed with everything you need to start creating amazing AI-generated videos and images. 🚀

UPDATES

01/03/2025

  • Create a cool little Gallery tool for Jupyter Labs (ManageGallery.ipynb) which can view all the created videos which can you multi select and delete. It will be avaliable on the output directory. Open the file and then ▶️

  • Add all the 1.5 Workflows to the AllinOne2.5 template

26/02/25

  • Ultra 1.4 is now is on the CUDA2.5 Template

  • Using Cuda 2.5 gives you more choices of GPU so will update this version for now until I see better compatablity for Cuda 2.6

25/02/25

  • Fixed many of the bugs but the Easy Setnode (Modified version works great that comes with this template)

  • Created CUDA 2.5 version to apply for more GPU and comptiable with Novita Cloud services. (11.8 image used)

  • Fix Permissions download-files.sh Unix bug

  • libEGL.so not loaded fix for img2vid

  • Added all nodes requied for 1.4

  • JupyterLab Password in the Runpod ENV template as "mypassword"

21/02/25

  • Added download-files.sh - Add your models to files.txt and './download-files.sh' on terminal to download all your files from hugginface

  • Updated Python to comfy recommened 3.12 - Uses Miniconda

20/02/25

  • Everything working well now udpated to the latest Pytorch and Cuda (Better performance)

  • One anoying issue is the Easy Setnode refuse to work so I've created an alternative AllinoneUltra1.3 without that node


Hunyuan Video - ComfyUI VScode 💥 AllInOne Template Guide

This guide will walk you through setting up and running the Hunyuan Video Generation environment on RunPod.


https://www.runpod.io/console/explore/l7ynpsgj7l

Prerequisites

Step 1: Setting Up Your RunPod Environment

  1. Log in to your RunPod account

  2. Navigate to the "Pods" section

  3. Click "Deploy" to create a new pod

Step 2: Selecting Template

  1. In the template selection:

    • Select "Custom Template"

    • Container Image: Hunyuan Video - ComfyUI Manager AllInOne3.0-CUDA2.5

    • Choose a GPU (Recommended: RTX 4090 or better)

    • Select at least 24GB RAM

    • Storage: Minimum 60GB (Recommended: 100GB)

    • CUDA 12.4 Minimum for 3.0-CUDA2.5



Step 3: Port Configuration

The following ports need to be exposed:

  • 8188: ComfyUI web interface

  • 8888: Jupyter Lab

These are pre-configured in the template, but verify they're exposed in the RunPod UI.

Step 4: Deploying Your Pod

  1. Click "Deploy" to start your pod


    About 25mins to download all the models. Will try to improve this with parallel downloads.


    READY TO GO

Step 5: Accessing the Interfaces

After deployment, you can access:

  1. ComfyUI Interface:

    • Click on "Connect" in your pod's details

    • Select port 8188

    • This opens the ComfyUI web interface (Not availiable until models are downloaded)

  2. Jupyter Lab is availiable from the start and can download models/loras

    • Click on "Connect"

    • Select port 8888

Wait for the green light!
After the models are downloaded the /workspace/ComfyUI folder will appear on the workspace directory and 8188 will go green ready to use. Juypter is avaliable even when its orange

JUPYTER LAB is avaliable right away.
- can download model's and loras and move them to the relevent ComfyUI folder when ready

Step 6: Using the Environment

ComfyUI Workflow

  1. Navigate to the ComfyUI interface

  2. Load the provided workflow:

    • Click the folder icon in the top-right

    • Select "AllinOneUltra1.3.json" from the workflows folder. This is slighly different to the original as the Easy Setnode is not working on the current setup and using KJNodes version


📥Step 7: Downloading Models

From HuggingFace 🤗 or Civitai or Google Drive🎨

  1. Open JupyterLab terminal and navigate to workspace and edit files.txt and run download-files.sh:

cd /workspace/files.sh
cd /workspace/download-files.sh



Models Used

The following models are automatically downloaded on first startup:

  • hunyuan_video_720_cfgdistill_bf16.safetensors (UNET)

  • Long-ViT-L-14-GmP-SAE-TE-only.safetensors (Text Encoder)

  • llava_llama3_fp8_scaled.safetensors (Text Encoder)

  • hunyuan_video_vae_bf16.safetensors (VAE)

  • clip-vit-large-patch14.safetensors (CLIP Vision)

  • hunyuan_video_FastVideo_720_fp8_e4m3fn.safetensors (FastVideo LoRA)


Troubleshooting

Common Issues

  1. Easy Setnode not working

  2. Comfyui not building correctly - Have to start again and create new pod

Access Terminal

Tips: Enable Preview

🤝 Community Support

If you need help or want to share your experiences, Your feedback helps improve the template for everyone!

Maintenance

Updating the Environment

To update to the latest version:

  1. Stop your current pod

  2. Delete the pod (your workspace volume will be preserved)

  3. Deploy a new pod using the latest image

Backup

Important files are stored in:

  • /workspace/ComfyUI/models/ - Model files

  • /workspace/ComfyUI/user/default/workflows/ - Workflows

  • /workspace/ComfyUI/output/

Support

For issues or questions:

  1. Check the GitHub repository issues section

  2. Check RunPod documentation for platform-specific issues

Additional Resources

BUGS & ISSUES ( Allinone3.0)

Any errors: ( i.e Node Type: SamplerCustomAdvanced - Exception Type: subprocess.CalledProcessError)

  • Disable the 3 nodes on Speed Tricks

BUGS & ISSUES (OLD Allinone Template)


RuntimeError: No CUDA GPUs are available

  • Only have NVIDIA_VISIBLE_DEVICES=all in ENV and remove CUDA_VISIBLE_DEVICES=all

BUILD Issue - [FATAL tini (19)] exec /workspace/start.sh failed: No such file or directory

  • This seems to happen when I try the A40 on Secure Cloud. Trying to fix this issue ( seems to work on community cloud)

Header Too Small - Download issues

  • I've added a downlaod-fix.sh file to the ./workform directory that will check the size of the safetensors files and if the size is too small (<10mb) then to re-download those files.

  • Run the file by navigating to the folder on terminal and' ./downlaod-fix.sh'

  • if it fails you may need to give it permission 'chmod +x ./downlaod-fix.sh'

VSCode Setup Issues

  • VSCode might not set up correctly on first try

  • Solutions:

    1. Rebuild the container

    2. Use SSH terminal to transfer files

    • Note: Community Cloud GPUs seem to have fewer issues with VSCode setup

VAE Issues

  • If you encounter: "VAELoader Error while deserializing header: HeaderTooSmall"

    • Solution: The latest image has been updated with Kijai's VAE which has better compatibility

    • There seems a be a few versions of the vae model out there but Kijai's one worked for me

Version Compatibility

  • Template works with both Ultra 1.2 and Ultra 1.3 workflows ( Support Added Wavespeed and added Triton)

Additional Notes

  • A comprehensive guide on usage will be published soon

  • Template is continuously being updated to fix GPU compatibility issues

  • For latest updates and changes, check the template description on RunPod



19

Comments