●  Setup guide · 15 minutes · once

Start your studiowithout a line of code.

Your computer does no computing: everything runs on a GPU rented from RunPod. You create a storage volume to keep your models and images, start a pod with the Lazarus template, then drive everything from Lazarus Control.

Get started Open the template ↗
A RunPod account$10 of creditA browser
  1. 00Create your RunPod account2 min
  2. 01Create the storage2 min
  3. 02Deploy the pod3 min
  4. 03Install the packfirst time
  5. 04Open Lazarus Control1 min
  6. 05Stop the pod10 s
How it works

Three pieces.Only one to switch on.

The storage keeps everything, even with the pod off. The pod is the machine with the GPU: switch it on to work, off afterwards. Lazarus Control opens in your browser and sends the work to the pod.

YOUR BROWSER LAZARUS CONTROL · PC OR MAC FORMS IMAGES · VIDEOS THE RUNPOD POD GPU · COMFYUI · JUPYTER BILLED HOURLY WHILE ON WRITES READS STORAGE KEEPS EVERYTHING
00

Create your RunPod account

Sign up at runpod.io, then add credit under Billing. $10 is enough to start. RunPod bills by the second: you pay for the GPU only while a pod runs, plus the storage each month.

  • Starting credit $10
  • Billing per second
  • Payment card or crypto
What you will pay
WhatWhenObserved price
GPU RTX PRO 6000 · 96 GoOnly while the pod runs≈ $2.09 / h
GPU RTX 4090 · 24 GoOnly while the pod runs≈ $0.74 / h
Storage · 300 GBAll the time, even with the pod off≈ $21 / month

Prices observed on RunPod; they vary by datacenter and demand. Example: one hour of images on a PRO 6000 ≈ $2; a 3,000-step training ≈ 1 h 40 min, so ≈ $3.50.

01

Create the storage

The network storage (Network Volume) keeps your models, your LoRAs and all your images while the pod is off. Without it, everything is lost at each stop. You create it once.

  • Size 300 GB
  • Minimum 200 GB
  • Datacenter same as your GPU
Shortcut: Lazarus Control can create the storage and rent the pod on its own. All it needs is a RunPod API key (console.runpod.io → Settings → API keys). Then skip steps 01 and 02 and go straight to step 03.

Simplified preview: the real RunPod screen may differ slightly.

  1. In RunPod, open Storage → New Network Volume.
  2. Datacenter: pick one that offers RTX PRO 6000 GPUs (we use EU-RO-1). The pod must run in that same datacenter.
  3. Size: 300 GB. The models alone weigh over 100 GB, and you need room for your images, videos and LoRAs.
  4. Name: lazarus, then Create.
02

Deploy the pod with the template

The Lazarus template contains ready-to-use ComfyUI, Jupyter and the right ports. You only pick the GPU and your storage.

  • Template LAZARUS ComfyUI
  • ID ga53ushz3b
  • Type On-Demand
Deploy the template ↗
  1. Click Deploy the template: RunPod opens with the Lazarus template already selected.
  2. At the top of the page, select your "lazarus" Network Volume. The GPU list is then limited to its datacenter.
  3. Pick the GPU (table below).
  4. Edit Template → Environment Variables → add JUPYTER_PASSWORD with a password of your own. It is used to start training from Lazarus Control.
  5. Keep On-Demand, not Spot, then Deploy.
  6. Wait 2 to 3 minutes for the pod to show Running.
GPUWhat it runsPrice
RTX PRO 6000
96 GBRecommended		Everything: dataset, fast training, images, face swap, motion control≈ $2.09/h
RTX 4090
24 GB		Dataset, images, face swap, slow training (24 GB mode). No motion control.≈ $0.74/h
Spot = cheaper but cut off without warning. For training or a video, stay On-Demand.
03

One button, and it is ready

Two lines to paste, one button: the panel installs the nodes, downloads the models, copies the workflows and starts ComfyUI. Done once, on a fresh storage volume.

  • Time about 20 min
  • Download 100 GB+
  • How often once
  1. On your PC: unzip the pack and double-click start.bat. Lazarus Control opens in your browser.
  2. Pick Create my pod for me and paste your RunPod API key — or I already have a pod with its id and Jupyter password — then press the button.
  3. One model needs a free Hugging Face account: open huggingface.co/black-forest-labs/FLUX.2-klein-9B, accept the licence, then Settings → Access Tokens to create a read token and paste it into the panel field. Without it the dataset generator and the face swap will not start.
  4. Leave the window open: the list ticks itself off (connect, inventory, install, start, final check). About twenty minutes the first time.
  5. When everything is green, click Create my first images.
A red line tells you what to do: pod stopped, password refused, missing model link. Press the button as often as you like, it only installs what is missing.
04

Open Lazarus Control

Lazarus Control runs right on your pod. Paste your pod ID (shown in RunPod, or the full address): you get your links.

  • Address <id>-8188.proxy.runpod.net/lazarus/
  • Server detected automatically
  • Data stays in your browser
Your pod
  1. Open Lazarus Control: your pod's server is detected automatically. The home screen shows the 5 steps and what is ready.
  2. Follow the order: dataset, training, then images, face swap and video with your new model.
  3. For training, open Lazarus Control from your PC: unzip the pack, double-click start.bat, go to Train the model and enter the pod ID and your Jupyter password.
01Dataset2 photos of your character → a training set.
02TrainingYour character's LoRA, with its trigger word.
03ImagesYour character in any scene.
04Face swapYour character on the photo of your choice.
05MotionA TikTok video, the same moves, 60 fps.
05

Stop the pod

The GPU is billed as long as the pod runs, even when idle. When you're done: Stop. Your files stay in the storage.

ButtonWhat happensTo resume
StopThe GPU stops, nothing is billed by the hour anymore. Nothing is lost.Start the same pod (if the GPU is free).
TerminateThe pod is deleted. The storage and everything in it remain.Redeploy the template with the same storage: everything is there.
Delete the storageEverything is erased: models, LoRAs, images.Only if you quit for good.
GPU unavailable when restarting? Terminate, then redeploy the template on the same storage with another GPU in the datacenter. Your files are in the storage, not in the pod.
Troubleshooting

Not working?Almost always one of these seven things.

Lazarus Control shows a 404 error or does not respond

The pod is stopped or still starting. Wait until it shows Running in RunPod. RunPod sometimes stops a pod on its own (status Exited): restart it, your files are in the storage.

Lazarus reports "missing nodes"

The install has not finished, or ComfyUI has not restarted since. Run the step 03 command again, then Restart the pod.

A style LoRA is reported as missing

Four style LoRAs (Detailer-KREA2, RealisticSnapshotKrea2, Famegrid-Natural-V1-Krea-2, phone_photography_2020_krea2) are community LoRAs and are not part of the pack. Lazarus switches them off on its own and the image is still made. To match the examples exactly, download them yourself and drop them into models/loras.

No GPU available

Your storage sets the datacenter. Pick another GPU in that datacenter, or try again a bit later: availability changes fast.

"CUDA out of memory" error

Motion control and fast training need the RTX PRO 6000. On a 4090, pick the "24 GB" mode in Train the model and skip video.

Jupyter asks for a password

It is the one you set in JUPYTER_PASSWORD at step 02. Forgot it? Edit Pod in RunPod, change the variable, restart.

Training won't start from Lazarus Control

Open Lazarus Control from your PC with start.bat (not from the pod address), and install once pip install requests websocket-client. The "Check the pod" button tells you what is missing.