Skip to content

📘 Tutorial run-comfyui-ltx

  • This tutorial reflects my own experience on RunPod.
  • Always consult the excellent official runpod documentation.

🚀 Starting a Pod

🧩 Choose a Template

t2v/i2v template example

Example: 👉 RunPod LTX-2.3 dev i2v t2v vi2v vt2v bf16/fp8

Steps:

  1. Choose a GPU
  2. Edit template settings if needed.
  3. Choose Volume disk (/workspace)
  4. Enable Volume encryption if desired.
  5. Click Deploy On-Demand.

⚠️ CUDA errors when deploying pod

Filter CUDA

  • Deploy on an other region.
  • Change filter to CUDA 12.8 on runpod's console

📜 Viewing System Logs

deployment start

  • Go to Logs.
  • Loading takes 9–15 minutes depending on region.
  • If the image doesn’t begin downloading after 1 minute, delete and redeploy in another region.

extracting start

Ends with (example):

Download extracting end

🐳 Viewing Container Logs

start container

When you see:

final message

→ Your pod is ready.

🔌 Connecting to Your Pod

Docs

Possible http services

⚠️ Not ready – Make sure your service is running! or Browser unautorised error.

Unautorised

Try to Connect with the pod-id and the port number to the proxy. You can find the url’s at the end of the log file in the runpod's console.

  • ComfyUI: https:<pod-id>-8188.proxy.runpod.net/login
  • Code-Server: https:<pod-id>-9000.proxy.runpod.net/login

🎨 ComfyUI

  1. Select tab Connect → ComfyUI
  2. Set username/password
  3. Load a workflow
  4. Press Run
  5. Monitor GPU/RAM via Telemetry

Select Workflow template

⚠️ ComfyUI's screen remains blank

  • Wait one minute and try again.
  • Restart your browser and/or clear cache.
  • Try with another browser (brave, chrome, edge).

💻 Web Terminal

Web Terminal

  • Select tab Connect
  • Enable web terminal
  • Provides terminal access directly in your browser.

🧑‍💻 Code-Server Login

Login without password variable set

No "PASSWORD" set

  • Copy the password displayed at the end of the container log file of the Runpod console or open the web terminal and enter.
cat /root/.config/code-server/config.yaml

Copy the password → log in via the Code-Server service on tab Connect.

Information in pod available

Code-Server

⚠️ Code-Server's screen remains blank

  • Wait one minute and try again.
  • Restart your browser and/or clear cache.
  • Try another browser (brave, chrome, edge).

"PASSWORD" set as env in Runpod template

Log in via the Code-Server service on tab Connect.

🔐 Secrets

Docs

Useful secrets:

  • PASSWORD (code-server login)
  • CIVITAI_TOKEN (civitai)
  • HF_TOKEN (huggingface)

📥 Downloading Models and LoRAs

From web terminal, Code-Server or ComfyUI-Lora-Manager.

🧩 ComfyUI-Lora-Manager

Launch web interface

top_bar_comfyui

  • Topbar ComfyUI.
  • Url displayed at end of container log file.
https:<pod-id>-8188.proxy.runpod.net/loras

Civitai token (needed for download)

  • Go to preferences and add your token if not set before starting the pod (CIVITAI_TOKEN).

Topbar

top_bar

  • Press Refresh and Fetch to download images for lora's available in the pod.
  • Press Download and add the civitai's URL (not download link).

Integration basic

  • Add node Lora-Loader (LoraManager) to your ComfyUI workflow.
  • Press the Paper Airplane in the Lora-Manager web interface.
  • Your lora is available in your workflow.

send_lora

lora_loader

🧩 Model linker

  • Tries to link models found in workflows to the provisioning in the pod.
  • ⚠️ Can download models but be aware of the limited volume capacity of the pod.

Example author

Example linking models in a ComfyUI template.

ml1

ml2

ml3

ml4

ml5

🧩 CivitAI CLI

If no "CIVITAI_TOKEN" was set, create or use a free token from the civitai website.

civitai version id

export CIVITAI_TOKEN=“xxxxx”
civitai_com  VERSION_ID /workspace/ComfyUI/models/<loras, etc>
civitai_red  VERSION_ID /workspace/ComfyUI/models/<loras, etc>

civitai_com 2228466 /workspace/ComfyUI/models/loras/
civitai_red 2893442 /workspace/ComfyUI/models/loras/

Refresh

Refresh ComfyUI pressing key r.

☁️ HuggingFace cli

"HF_TOKEN" is mandatory but needed for gated sites or upload.

Login:

hf auth login --token xxxxx

or set token:

export HF_TOKEN="xxxxx"

Download example from Huggingface.

hf download ricecake/wan21NSFWClipVisionH_v10 wan21NSFWClipVisionH_v10.safetensors --local-dir /workspace/ComfyUI/models/clip_vision

Refresh ComfyUI pressing key r.

Manual provisioning

  • Information is available in the pod's documentation.
  • Open web terminal or code-server.

🧩 Deleting Models and LoRAs

Web Console or code-server

  • Type "ncdu" in the console
  • Follow the instructions.

Lora Manager

  • Select lora
  • Select delete.

🧩 ComfyUI-manager

  • Topbar or menu

manage_extensions

🧩 Restart pod

  • To restart ComfyUI restart the pod from runpod’s console (no information loss).

Restart_pod

🧩 Stop and restart a pod

  • No information loss as ComfyUI is copied to the /workspace volume.
  • Use this option to pause your pod.

⚠️ Be aware it is possible that no GPU is available when you restart the pod.

  • Before stopping the pod copy your creations to cloud/local.
  • You can still access the pod from with ssh or web console (0.5 vCPU and very little memory).

🔄 Uploading & Downloading Files

☁️ Cloud Sync

Free dropbox

Cloud sync

  • Reliable and fast upload and download for large files.
  • Go to Dropbox developers
  • Create an app to connect with runpod.io.

📦 runpodctl

Speed

  • Fine for downloading files from your pod.

  • Problematic for uploading large files to your pod with slow connections (timeouts).

  • You need to install a client on your local computer.

  • You do not need an api key to upload/download to/from your pod.

Install windows

Invoke-WebRequest "https://github.com/runpod/runpodctl/releases/latest/download/runpodctl-windows-amd64.exe" -OutFile "runpodctl.exe"

install linux

wget --quiet --show-progress https://github.com/runpod/runpodctl/releases/latest/download/runpodctl-linux-amd64 -O runpodctl && chmod +x runpodctl && sudo cp runpodctl /usr/bin/runpodctl

🔧 Advanced Features

🔐 SSH / SCP / SFTP

Docs.

Key generation

  • Linux has ssh-keygen already installed to generate keys.
  • Windows install Git to get ssh-keygen.
  • Put your public key on runpod.io keep your private key on your local computer.

Apps.

🔑 RunPod API

Useful when the web console is out (examples are outages from AWS and Cloudflare).

Using runpod.io API

  • Docs.
  • Install with "sudo apt install curl" on Ubuntu.
curl --request GET \
  --url https://rest.runpod.io/v1/pods \
  --header 'Authorization: Bearer <api-key>'

curl --request POST \
  --url https://rest.runpod.io/v1/pods/{podId}/stop \
  --header 'Authorization: Bearer <api-key>'

curl --request DELETE \
  --url https://rest.runpod.io/v1/pods/{podId} \
  --header 'Authorization: Bearer <api-key>'

Using runpodctl to start/stop your pod with your api key.

  • Docs.
  • Install client on local computer.
  • Use PowerShell or Bash.

Usage

runpodctl doctor

runpodctl pod list                 
runpodctl pod stop <pod_id> 
runpodctl pod delete <pod_id>