Video Generation with Wan 2.2 and ComfyUI

This guide walks through generating video from text prompts using the Wan 2.2 model inside a ComfyUI workflow on Laboratory OS.

Prerequisites

• Laboratory OS running with GPU access
• Minimum 8 GB VRAM (16 GB recommended for 720p output)

Step 1: Install ComfyUI

Open the Apps panel from the Laboratory OS desktop and find ComfyUI. Click Install. The app downloads and starts automatically — this takes a few minutes on first run.

Once installed, ComfyUI is available at its own subdomain:

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

A shortcut also appears on your desktop.

Step 2: Download the Wan 2.2 Model

Open the Model Library from the Laboratory OS desktop. Search for Wan 2.2 and click Download on the variant appropriate for your VRAM:

Models download directly into the container’s shared model directory and become available to ComfyUI immediately — no restart needed.

Step 3: Load the Text-to-Video Workflow

Open ComfyUI from your desktop shortcut. In the menu bar, go to Workflow → Browse Templates and select the Wan 2.2 Text-to-Video template.

The workflow loads with these key nodes:

• CLIP Text Encode — your prompt
• WanVideoSampler — the core diffusion sampler
• Video Combine — assembles frames into an output file

Step 4: Configure Your Prompt

Click the CLIP Text Encode node (positive prompt). Write a descriptive prompt:

A time-lapse of a storm rolling over a mountain range, cinematic, 4K, dramatic lighting

For the negative prompt node, add:

blurry, low quality, watermark, text, distorted

Step 5: Set Sampling Parameters

In the WanVideoSampler node, recommended starting values:

Tip: Lower frame counts render faster. Start with 49 frames (≈ 2s) for quick iteration.

Step 6: Generate

Click Queue Prompt in the top-right. ComfyUI shows progress in the bottom status bar. On a 16 GB GPU, a 3-second 480p clip takes roughly 2–4 minutes.

The finished video appears in the output panel and is saved to /workspace/ComfyUI/output/ inside the container.

Accessing Your Output

Open the File Browser app from the Laboratory OS desktop and navigate to:

/workspace/ComfyUI/output/

Files can be downloaded directly from the browser interface.

Performance Tips

• Reduce steps to 15 for faster drafts with similar quality.
• Enable fp8 quantization in the model loader node to cut VRAM usage by ~30%.
• Use the Video Combine node’s frame_rate setting to control playback speed without re-rendering.

What’s Next

• LLM Hosting with Qwen3 and vLLM — add a local AI backbone to your creative pipeline
• ComfyUI on Laboratory OS — image generation workflows