Why is Wan2.2 not generating images in ComfyUI?

Check that all model files are downloaded and placed in the correct folders. Also make sure each file is selected in the correct workflow node. Refresh your browser after adding new files to ComfyUI.

Is Wan2.2 free to use locally?

Yes. All Wan2.2 model weights are available for free on Hugging Face and can be run locally inside ComfyUI without any subscription or API cost.

Install Wan2.2 Text-to-Image in ComfyUI (2026 Step-by-Step Guide)

Q: What is Wan2.2 Text-to-Image?

Wan2.2 is a powerful AI model that converts text prompts into detailed, high-quality images. It runs locally in ComfyUI using a set of downloadable model files including a diffusion model, VAE, text encoder, and LoRA.

Looking to create stunning images from your text prompts using the latest AI technology? The Wan2.2 text-to-image model makes it simple to generate high-quality visuals right on your own computer. This guide covers how to install Wan2.2 Text-to-Image locally with ComfyUI, including all downloads, model file setup, and step-by-step instructions for beginners.

What is Wan2.2 Text-to-Image?

Quick answer: Wan2.2 is a powerful 14B-parameter AI model that converts text prompts into detailed, high-quality images. It runs fully locally in ComfyUI — no cloud subscription or API key required.

Wan2.2 is a powerful AI model that converts text into detailed images. It works with ComfyUI, a node-based interface, letting anyone generate creative images by typing a prompt. No advanced AI knowledge needed — just set up the files and run the workflow.

The model uses a dual-diffusion architecture with a high-noise and low-noise model working in tandem, guided by the UMT5 XXL text encoder and a dedicated VAE. A LoRA is also included to accelerate generation with fewer steps.

Download Workflow

Quickly get started by downloading the ready-to-use workflow file for ComfyUI. This file is pre-configured with all the correct node connections — just load it and select your model files.

💡 Tip: Load the workflow into ComfyUI first, then download and place the model files. The workflow will show you exactly which node needs which file.

Drag and drop this JSON file into your ComfyUI workspace to load it.

Required Models

Before running the workflow, download the following model files and place them in the correct folders inside your ComfyUI directory.

Diffusion Models

Folder Structure — Where to Place Model Files

Place the downloaded files into these folders inside your ComfyUI directory:

ComfyUI/
└── models/
    ├── diffusion_models/
    │   ├── wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors
    │   └── wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors
    ├── vae/
    │   └── wan_2.1_vae.safetensors
    ├── text_encoders/
    │   └── umt5_xxl_fp8_e4m3fn_scaled.safetensors
    └── loras/
        └── lightx2v_14B_T2V_cfg_step_distill_lora_adaptive_rank_quantile_0.15_bf16.safetensors

⚠️ Warning: Placing a file in the wrong folder is the most common setup error. Double-check each file is in the exact folder shown above before running the workflow.

Step-by-Step Setup Guide

Follow these five steps to get Wan2.2 Text-to-Image running inside ComfyUI on your local machine.

Update ComfyUI

Make sure ComfyUI is up to date using its built-in manager or by pulling the latest commit. New models often require recent node versions.

Import the Workflow

Drag and drop the downloaded JSON workflow file into your ComfyUI canvas to load it. All nodes will be pre-connected.

Load the Models

In the two Load Diffusion Model nodes, select wan2.2_t2v_high_noise_14B_fp8_scaled.safetensors and wan2.2_t2v_low_noise_14B_fp8_scaled.safetensors respectively.

In the Load CLIP node, select umt5_xxl_fp8_e4m3fn_scaled.safetensors.
In the Load VAE node, select wan_2.1_vae.safetensors.
In the Load LoRA node, select the lightx2v_14B_T2V_... file.

Enter Your Prompt

Type your desired scene or subject in the prompt box. See the example prompts in the next section for inspiration. Adjust image width, height, steps, or other settings as needed.

Generate the Image

Click Queue Prompt / Run to create your image. Wait for the process to complete — generation time depends on your GPU and selected resolution.

Example Runs

Here are three real outputs generated with Wan2.2 Text-to-Image in ComfyUI. Each card shows the exact prompt used and the resulting image — click any image to view it full size.

Example 1 — Abstract Expressionism — Prompt Used

Bask in the realm of abstract expressionism as a mysterious figure cloaked in a vibrant red shawl steps into view against an ebony void. Her head hangs low, enveloped in a cloud of bewitching curiosity. In stark contrast with her crimson surroundings, crisp white lines dance alongside her form, accentuating each delicate curve as if painted by an ethereal brushstroke. A sense of profound contemplation radiates from within as she stands on the edge of a transformative moment, encapsulated in a timeless swirl of emotions unraveling before our eyes.

Wan2.2 ComfyUI output — abstract expressionism figure in vibrant red shawl against dark background with flowing white lines — A crisp, magazine-like portrait with professional studio lighting and enhanced fabric textures.

Example 2 — Realistic Street Scene — Prompt Used

A realistic young woman standing on a narrow European street just after heavy rain. The cobblestones are slick and mirror-like, reflecting neon signs and warm cafe lights. She wears a fitted leather jacket with visible creases and natural wear, paired with light denim jeans showing subtle stitching details. Her face has soft skin textures and natural pores, lit by diffused streetlamp glow. Small puddles ripple from light raindrops, each droplet rendered in sharp realism.

Wan2.2 ComfyUI output — realistic young woman on wet European cobblestone street with neon reflections and cinematic lighting — A vivid, cinematic scene with rich atmospheric lighting and realistic wet-street textures.

Example 3 — 3D CGI Doberman — Prompt Used

(best quality, 8K, high resolution, masterpiece), ultra detailed, (3D CGI), black sunglasses, trendy, fashionable, silver and black stylized angry muscle doberman on a black background, countryside advertising, winning photo.

Wan2.2 ComfyUI output — 3D CGI stylized silver and black angry muscle doberman with sunglasses on dark background — A high-detail 3D CGI render with sharp product-advertising composition and dramatic lighting.

Tips & Troubleshooting

If something isn't working as expected, one of these is almost always the cause:

Images not generating — model files missing or in wrong folder

Fix: Double-check that all five model files are downloaded and placed in the exact correct folders (diffusion_models, vae, text_encoders, loras) and selected in the matching workflow nodes.

Slow generation performance

Fix: Lower the image resolution or reduce the number of sampling steps. The fp8 model variants are already optimized — ensure you downloaded the fp8 versions and not the full bf16 files.

Results not matching expectations

Fix: Use more descriptive, detailed prompts. Increase the CFG scale for stronger prompt adherence, or try adjusting the step count upward for more refined outputs.

Out-of-memory errors (CUDA OOM)

Fix: Reduce image resolution first, then lower the batch size. If you're still hitting VRAM limits, make sure you're using the fp8 model files which require significantly less VRAM than full-precision variants.

New model files not appearing in node dropdowns

Fix: After placing files in ComfyUI folders, press F5 or Ctrl+Shift+R to fully refresh ComfyUI in your browser. Without a full refresh, new files won't appear in the node dropdowns.

💡 Tip: For a complete list of ComfyUI errors and fixes, see the ComfyUI Troubleshooting Guide.

Final Thoughts

The Wan2.2 text-to-image workflow brings advanced AI image generation to your desktop, making creative projects easier and more accessible. With this easy local setup, anyone can install Wan2.2 in ComfyUI and start producing beautiful or imaginative images from simple text descriptions. The fp8 model files keep VRAM requirements manageable for mid-range hardware. Give it a try and unlock new ways to bring your ideas to life!

Frequently Asked Questions

Wan2.2 is a powerful 14B-parameter AI model that converts text prompts into detailed, high-quality images. It runs fully locally in ComfyUI using a dual-diffusion architecture guided by the UMT5 XXL text encoder.

Place the two diffusion model files in ComfyUI/models/diffusion_models, the VAE in ComfyUI/models/vae, the text encoder in ComfyUI/models/text_encoders, and the LoRA in ComfyUI/models/loras.

Check that all five model files are downloaded and placed in the correct folders. Make sure each file is selected in the correct workflow node. Also refresh your browser after adding new files so ComfyUI picks them up.

Lower the image resolution or reduce the number of sampling steps. The fp8 model files (already recommended here) significantly reduce VRAM usage and generation time compared to full-precision variants.

Yes. All Wan2.2 model weights are available for free on Hugging Face and can be run locally in ComfyUI without any subscription, API key, or ongoing cost.