Image Generation with ComfyUI

This guide walks through generating your first image with ComfyUI on Open Laboratory, from a fresh container to a finished image.

Prerequisites

A running Open Laboratory container with an NVIDIA GPU
Your desktop open at https://{your-slug}.tunnels.laboratory.computer

If you haven’t started a container yet, see the Quick Start.

Step 1: Install ComfyUI

From the Open Laboratory desktop, open the Apps panel and find ComfyUI. Click Install.

ComfyUI will download its dependencies and set up its environment inside the container. This takes 2–4 minutes on first install.

Once installed, click Launch. ComfyUI opens in a new browser tab at:

https://{your-slug}--comfyui.tunnels.laboratory.computer

Step 2: Download a Model

ComfyUI needs a Stable Diffusion checkpoint to generate images. Switch back to your desktop and open the Models panel.

A good starting point is SDXL Base 1.0 — it produces high-quality results and has broad community support. Click Download.

The model downloads directly into the container at /data/models/checkpoints/. This typically takes 2–5 minutes depending on your host’s internet connection.

Step 3: Load the Default Workflow

Switch back to the ComfyUI tab. If it’s showing an empty canvas, load the default workflow:

Click the Load button in the right panel (or drag a workflow JSON onto the canvas)
Choose the built-in default_workflow.json

The default workflow is a simple text-to-image pipeline: CLIP Text Encode → KSampler → VAE Decode → Save Image.

Step 4: Select Your Model

In the workflow, find the Load Checkpoint node. Click the model dropdown and select the checkpoint you downloaded (e.g., sd_xl_base_1.0.safetensors).

Step 5: Write a Prompt

Find the CLIP Text Encode (Prompt) node connected to the positive conditioning input. Click inside it and write your prompt:

a photograph of a mountain lake at sunrise, golden hour, cinematic, 8k

In the negative prompt node, add things you want to avoid:

blurry, low quality, oversaturated, watermark

Step 6: Generate

Click Queue Prompt in the top-right corner. ComfyUI queues the job and starts generating. You’ll see a progress bar on the KSampler node.

On a modern GPU (RTX 4090, A100), SDXL at 1024×1024 takes about 15–30 seconds with 20 steps.

Your finished image appears in the canvas and is saved to /data/apps/comfyui/output/ inside the container.

Tips

Faster generation: Reduce steps to 12–15. Use a Lightning or Turbo SDXL LoRA for near-instant results at slightly lower quality.

Higher quality: Increase steps to 30–40. Use a DPM++ 2M Karras sampler.

Changing resolution: Edit the Empty Latent Image node. SDXL is trained for 1024×1024 but works well at other resolutions too — try 1152×896 for landscape.

Using LoRAs: Add a Load LoRA node between the checkpoint loader and the KSampler. Download LoRAs from the Models panel and stack multiple at once with adjustable weights.

Saving workflows: Click Save in the right panel to export your workflow as JSON. You can reload it later or share it with others.