Workflows/Create images on a rented GPU with SDXL Turbo

Images / WORKFLOW

Create images on a rented GPU with SDXL Turbo

Rent one GPU, load the official SDXL Turbo graph in ComfyUI, generate and download an image, then stop or remove the rental. Your own machine only controls the browser.

EstimatedSource checked 9/29/2026

RENTED GPU

Freshness 2 out of 100. Repository age: 1,053 days old, Nov 27, 2023. Measures age, not quality.

Popularity 81 out of 100. 815,964 downloads in 30 days; 2,649 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.

What do these scores mean?

Two ways to compare models. Neither score measures answer quality, benchmark performance, or whether a model fits your machine.

Freshness · release age

A new release starts at 100. The score halves every 180 days: 100 × 2^(−age in days / 180). When the release date is unavailable, a repository date is a fallback—not proof of when the model was released. Missing dates stay unscored.

Check the date source

Popularity · adoption

Downloads over the past 30 days carry 80% of the score; lifetime likes carry 20%. Both use a logarithmic scale, capped at 10 million downloads and 100,000 likes. If likes are unavailable, downloads carry all the weight. Missing download statistics stay unscored.

round(80 × min(1, log10(1 + downloads) / 7) + 20 × min(1, log10(1 + likes) / 5))

Check the statistics source

Scores as of Oct 15, 2026. Statistics checked Oct 15, 2026. Statistics are marked for refresh after 72 hours.Full scoring method

Before you begin

Hardware

Choose the rented machine here. These rows describe remote GPU capacity, not the saved laptop or desktop.

Machine	SDXL Turbo · FP16
1× RTX 4090 · 24 GB VRAM + 32 GB RAM	Recommended; official standard ComfyUI template
1× L40 / L40S · 48 GB VRAM + 32 GB RAM	More headroom; compare the current hourly rate
1× RTX 5090 · 32 GB VRAM + 32 GB RAM	Use the official ComfyUI CUDA 13 template

One GPU is enough. Start at 512×512, batch 1, one Turbo step. A larger card is an option, not a requirement. Remote host RAM and storage are separate from GPU VRAM.

RENTED GPU IMAGE GENERATIONSOURCE-LINKED SETUP

Create images on a rented GPU with SDXL Turbo

Use ComfyUI through your browser while a Runpod GPU does the work. This complete route includes the model, loadable graph, private access, exported image and clear shutdown steps.

Choose a package Load the visual workflow Check the result

BEFORE YOU BEGIN

The documented package

This setup uses the package documented for this task. Review its source and supported platforms before starting.

Save your machine in My Hardware to get an automatic starting choice. You can always choose any package yourself.

CURRENT SETUP

SDXL Turbo · FP16 · official

Rent a 24 GB RTX 4090 with 32 GB RAM for comfortable headroom. The small 512×512, batch 1 Turbo workload can fit smaller GPUs, but this guide recommends a straightforward supported rental. Local-machine memory does not determine fit.

Freshness 2 out of 100. Repository age: 1,053 days old, Nov 27, 2023. Measures age, not quality.

Popularity 81 out of 100. 815,964 downloads in 30 days; 2,649 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.

What do these scores mean?

Two ways to compare models. Neither score measures answer quality, benchmark performance, or whether a model fits your machine.

Freshness · release age

Check the date source

Popularity · adoption

round(80 × min(1, log10(1 + downloads) / 7) + 20 × min(1, log10(1 + likes) / 5))

Check the statistics source

Scores as of Oct 15, 2026. Statistics checked Oct 15, 2026. Statistics are marked for refresh after 72 hours.Full scoring method

View weight source

THE WORKFLOW

Set up SDXL Turbo · FP16 on your machine

Choose where you want to run it. Follow the visual setup and import the linked workflow.

CHOOSE YOUR SETUPLinux

Linux execution host. Control it from Windows, Mac or Linux using a browser and authenticated SSH tunnel; no local GPU is required.

Choose a rental for this image setup

Use one RTX 4090 with 24 GB VRAM and at least 32 GB system RAM for this example. An L40/L40S is also suitable; an RTX 5090 needs the separate CUDA 13 template in the next step. Your laptop or Mac only needs a browser, an SSH client and room for downloaded images—the model runs on the rented Linux machine. Review the current On-Demand hourly rate and storage charges in Runpod before deploying. This first run is 512×512, batch 1, one sampling step. Read the SDXL Turbo model license before commercial use.

Open Runpod · affiliate Current billing rules Publisher model and license

Prepare an SSH key on your own computer

Use an existing SSH key or create a dedicated one. On Windows, open PowerShell; on Mac/Linux, open Terminal. Run ssh-keygen -t ed25519 and follow the prompts, using a new filename if the default key already exists. Add only the resulting .pub file contents to Runpod Account → Settings → SSH Public Keys. Keep the private key on your computer. Windows needs the OpenSSH Client optional feature if ssh is not recognized. Runpod’s linked instructions show the complete key setup. This key authenticates the private browser tunnel; a Runpod API key is unnecessary.

Runpod SSH key setup Windows OpenSSH client installation

Install the official ComfyUI template privately

In Runpod choose the linked official ComfyUI template for an RTX 4090 or L40/L40S. For a 5090 choose the linked ComfyUI CUDA 13 template instead. The template installs ComfyUI and required software on the Pod. In the deployment configuration select one GPU, On-Demand, a public IP for direct SSH, 50 GB container disk and 100 GB volume disk mounted at /workspace. Before deploying, clear the exposed HTTP ports and expose only TCP port 22. Do not publicly expose 8188, 8080 or 8888: the stock ComfyUI page has no login. The SSH tunnel below supplies authenticated access from your computer.

Official standard ComfyUI template Official ComfyUI CUDA 13 template Current template source and directory layout Runpod port exposure and authentication

Deploy and wait for the application to start

Confirm the displayed rate and click Deploy On-Demand. Open Pod → Logs and wait for ComfyUI to finish startup. First startup can take several minutes. In Connect, start the Runpod Web Terminal and open it. This is a Linux Terminal running on the rented machine even if your own computer uses Windows or macOS. Do not install ComfyUI again or replace the template’s PyTorch build.

Open the Pod terminal

Open ComfyUI through your private browser tunnel

In the Pod’s Connect panel, copy the command under SSH over exposed TCP. It contains this Pod’s real IP, mapped SSH port and your key path. On your own computer—not in the Pod Terminal—paste it into PowerShell on Windows or Terminal on Mac/Linux. Append -N -L 127.0.0.1:8188:127.0.0.1:8188 to that copied command, then run it. Keep this terminal open; no output while connected is normal. Open http://127.0.0.1:8188 in your local browser. If port 8188 is already used locally, change the first 8188 to 8190 and browse http://127.0.0.1:8190. The second port remains 8188. All generation still runs on the Pod.

Direct SSH connection details

Put the single checkpoint in its exact folder

In the Pod Web Terminal, run the block below. It downloads the official 6.94 GB FP16 checkpoint into the persistent ComfyUI workspace and checks its published SHA256. This all-in-one checkpoint supplies the model, text encoders and VAE; no separate encoder or VAE download is needed. Keep the Terminal open until the checksum says OK. The model file belongs on the Pod, not your laptop. The tree below shows its exact destination.

Put these files in /workspace/runpod-slim/ComfyUI/1 required file

models/checkpoints/

sd_xl_turbo_1.0_fp16.safetensorsOfficial SDXL Turbo FP16 checkpoint · 6.94 GBGet file

Publisher file and checksum

cd /workspace/runpod-slim/ComfyUI
mkdir -p models/checkpoints
curl -fL --retry 3 'https://huggingface.co/stabilityai/sdxl-turbo/resolve/main/sd_xl_turbo_1.0_fp16.safetensors' -o models/checkpoints/sd_xl_turbo_1.0_fp16.safetensors
printf '%s  %s\n' 'e869ac7d6942cb327d68d5ed83a40447aadf20e0c3358d98b2cc9e270db0da26' 'models/checkpoints/sd_xl_turbo_1.0_fp16.safetensors' | sha256sum -c -

Choose the SDXL Turbo template or workflow file

Easiest route: in the Pod’s ComfyUI page, click Templates in the sidebar (also Workflow → Browse Workflow Templates), search for “SDXL Turbo” and open that exact template. It uses the same sd_xl_turbo_1.0_fp16.safetensors checkpoint, SDTurboScheduler and core nodes as this guide. Or download the official workflow JSON below to your own computer (use Save link as if it opens as text), then drag it onto the ComfyUI canvas or use Workflow → Open. The original PNG is another loadable copy: use Save image as and preserve its workflow metadata; a screenshot or recompressed copy may lose it. Use one route. Refresh the model list and select sd_xl_turbo_1.0_fp16.safetensors in Load Checkpoint. Apply the prompt, fixed seed and YouRunAI/SDXL-Turbo output prefix in the next step; template defaults differ. No extra custom-node downloads are needed. If the template is unavailable on this Pod, the JSON or original PNG works with the same core graph.

How to open ComfyUI Templates Download official SDXL Turbo workflow JSON Download official loadable workflow PNG Official ComfyUI SDXL Turbo example

Set a small, repeatable first image

In Empty Latent Image set width 512, height 512 and batch_size 1. Set SDTurboScheduler steps to 1, KSamplerSelect to euler_ancestral, and SamplerCustom cfg to 1 with add_noise enabled. Set the noise seed to 12345 and its after-generation control to fixed. In Save Image set filename_prefix to YouRunAI/SDXL-Turbo. Keep Auto Queue off. Replace the positive CLIP Text Encode prompt with: “A matte black water bottle on a warm stone pedestal, soft studio lighting, clean product photograph, no text.” The negative branch may remain connected; CFG 1 does not add negative-prompt guidance.

Generate and inspect the result

Click Run once. Watch the nodes progress until the image appears in Preview Image/Save Image. Confirm it is a 512×512 PNG and check that the bottle, lighting and background broadly follow the prompt. This setup is text-to-image; it does not edit a supplied reference photo. If the image is distorted, keep the same seed and change one prompt detail; you may also try 2–4 steps with the Turbo scheduler. Do not substitute a normal SDXL graph or large CFG value. No render-time claim is assumed for your rented host.

Download both your image and the reusable workflow

Right-click the generated image and choose Save Image to download it to your own computer. Save/export the ComfyUI workflow JSON too, naming it yourunai-sdxl-turbo.json. Open the downloaded PNG locally before ending the rental. On the Pod, Save Image places outputs under /workspace/runpod-slim/ComfyUI/output/YouRunAI. You can access them again through ComfyUI before stopping the Pod. Keep a local copy of important work rather than relying only on the rented disk.

Stop compute when you finish and resume later

After downloading your files, return to Runpod → Pods, click Stop and confirm the Pod status is stopped. Closing the browser or SSH terminal alone does not stop GPU billing. A stopped Pod releases compute but its volume disk still incurs storage charges. /workspace survives a stop; the container disk does not. To resume, click Start, confirm a GPU is allocated, wait for ComfyUI, copy the current direct-SSH connection details again, reopen the tunnel and load your saved workflow. The workspace checkpoint should remain.

Stop and restart behavior Storage persistence and charges

Remove a rental you no longer need

If you are finished permanently, first verify the local image and workflow copies open, then terminate this Pod from Runpod. Termination deletes its container and volume-disk data. A separately created network volume persists and continues charging until you delete that volume separately; this starter uses the Pod volume disk instead. Check Billing and the Pods/Storage lists to confirm no unwanted resource remains. Finally close the SSH tunnel terminal.

Terminate and data retention Billing and storage

SUCCESS CHECK · SDXL Turbo · FP16

An image saved locally and a rental you can account for

You can open the generated 512×512 PNG on your own computer, reload the saved workflow, and identify the exact checkpoint and seed. Runpod shows the Pod stopped or terminated when finished; any retained storage is visible and intentional.

Source-checked setup with estimated hardware guidance; no Pod was rented or image generated as part of this review. The stock ComfyUI template does not authenticate its public 8188 page, so this guide keeps HTTP ports closed and uses an SSH-key tunnel. Only Runpod is linked as the GPU rental provider.

Back to setup steps

When it doesn’t go to plan

The ComfyUI browser tab will not open

Wait for startup to finish in Pod logs, confirm the Pod has a GPU, and keep the local SSH tunnel open. Browse the local tunnel address, not a public 8188 proxy URL. Copy fresh direct-SSH details after a restart; the assigned port may change.

My SSH command asks for a password or rejects the key

Confirm the .pub contents are saved in Runpod account credentials and the command’s -i path points to the matching private key. Use direct SSH over exposed TCP from Connect. Do not paste a fingerprint or the private key into the account field.

The checkpoint does not appear or the checksum fails

Confirm the download was run in the Pod Terminal and completed under /workspace/runpod-slim/ComfyUI/models/checkpoints. Re-run that exact download if it was interrupted, then check SHA256 and refresh the ComfyUI model list. Do not select a partial file.

Blackwell / RTX 5090 reports CUDA errors

Use the current official CUDA 13 template linked in this guide for the 5090. The older tutorial’s Blackwell title/path can be stale. Preserve/download important outputs before recreating a Pod with the correct template; do not randomly replace PyTorch in the running image.

The image is noisy or poor quality

Load the exact official Turbo workflow, select the FP16 Turbo checkpoint, use 512×512 and CFG 1, and keep batch at 1. Check SDTurboScheduler is present. Try 2–4 steps only after the one-step route works.

Charges continue after closing the tab

Close/cancel generation, download the image and workflow, then Stop the Pod in Runpod. Stopped volume disks remain billable. To end retention charges, terminate the Pod after exporting what you need and remove any separate storage you intentionally created.

REFERENCE LIBRARY

Sources and files

Original instructions, model files, and compatibility notes behind this setup.

12 SOURCES

01Runpod ComfyUI walkthroughdocs.runpod.io 02Current standard template sourcegithub.com 03Current CUDA 13 template sourcegithub.com 04Official Turbo workflowcomfyanonymous.github.io 05Checkpoint, limitations and licensehuggingface.co 06Exact model file and checksumhuggingface.co 07Runpod SSH authenticationdocs.runpod.io 08Runpod public port behaviordocs.runpod.io 09Runpod stop / start / terminatedocs.runpod.io 10Runpod storage lifecycledocs.runpod.io 11Runpod billingdocs.runpod.io 12Official built-in SDXL Turbo templategithub.com

The model behind this workflow

SDXL Turbo

Keep exploring

Text

SETUP GUIDE

Qwen3.5 · 9B private chat setup

Source-backed package and platform-route plan for private local chat with Qwen3.5 · 9B. Preserves the official Ollama 9B package, a smaller 4B family fallback, and a clearly labeled community uncensored GGUF option.

Freshness 41 out of 100. Release age: 230 days old, Feb 27, 2026. Measures age, not quality.

Build a private coding assistant with Qwen3.8 27B

Run Qwen3.8 27B with Ollama, connect it to Cline in VS Code, and make one reviewable code change in a tiny local project.

Freshness 76 out of 100. Repository age: 71 days old, Aug 5, 2026. Measures age, not quality.

Arena Open Source Web coding: rank 10 of 55. Ranked model: qwen3.8-27b. Original model ranking, not a benchmark of this package. Export dated Oct 13, 2026. Source checked Oct 15, 2026. Human preference ranking, not hardware suitability.+3

SOFTWAREOllama + Visual Studio Code + Cline + Python

Create images locally with FLUX.2 Klein

Generate images from text prompts in ComfyUI with the fast distilled or more adjustable base 4B model. Save your results and reuse the visual workflow.

Freshness 35 out of 100. Release age: 273 days old, Jan 15, 2026. Measures age, not quality.

SOFTWAREComfyUI

LOCALRENTED GPU

Verified source