How to Upscale Images with Qwen Image 2.1 in ComfyUI

1. Introduction

Qwen Image 2.1 is the latest open-source release from the Qwen family — a compact yet powerful 7B-parameter unified model that seamlessly handles both text-to-image generation and advanced image editing. Built with just 32 Single-Stream DiT layers and clever optimizations like mixed-granularity attention and prefix KV cache reuse, it delivers impressive visual quality at a surprisingly low computational cost.

In this tutorial we focus on its most exciting capabilities for high-quality image upscaling and native alpha-channel support. Whether you want to upscale images while preserving transparency, cleanly extract subjects from backgrounds, or perform versatile edits with reference images and masks, Qwen Image 2.1 handles it all in one model. Even better, while upscaling or removing the background you can always make small targeted edits at the same time. By the end of this guide you’ll have a ready-to-use ComfyUI workflow that lets you upscale images with beautiful results and full control over alpha channels.

2. Requirements & Setup

Requirement 1: ComfyUI

You'll need ComfyUI running either locally or on a cloud GPU.

Local (Windows): 👉 How to Install ComfyUI Locally on Windows

Cloud GPU (RunPod): 👉 How to Run ComfyUI on RunPod with Network Volume

Requirement 2: Update ComfyUI

Keeping ComfyUI and custom nodes up to date ensures the latest Qwen Image 2.1 nodes and optimizations are available.

Windows Portable Users: Navigate to: ...\ComfyUI_windows_portable\update Double-click: update_comfyui.bat

RunPod / Linux Users:

1 cd /workspace/ComfyUI && git pull origin master && pip install -r requirements.txt && cd /workspace

1 cd /workspace/ComfyUI && git pull origin master && pip install -r requirements.txt && cd /workspace

You can also update via ComfyUI Manager. After updating, verify the new nodes appear in the node search.

⚠️ Note: Make sure you are using the latest ComfyUI version 0.36.0 — older builds may not include the correct native nodes required for this workflow.

⚠️ Note: Make sure you are using the latest ComfyUI version 0.36.0 — older builds may not include the correct native nodes required for this workflow.

Requirement 3: Download the Required Models

Download the following files and place them in the correct folders:

Requirement 4: Verify Your Folder Structure

Confirm the files are in the correct locations before loading the workflow.

1📂 ComfyUI/
2├── 📂 models/
3│   ├── 📂 diffusion_models/
4│   │   └── qwen_image_2.1_bf16.safetensors
5│   ├── 📂 vae/
6│   │   └── qwen_image_2.1_vae_bf16.safetensors
7│   └── 📂 text_encoders/
8│       └── qwen3vl_8b_bf16.safetensors

1📂 ComfyUI/
2├── 📂 models/
3│   ├── 📂 diffusion_models/
4│   │   └── qwen_image_2.1_bf16.safetensors
5│   ├── 📂 vae/
6│   │   └── qwen_image_2.1_vae_bf16.safetensors
7│   └── 📂 text_encoders/
8│       └── qwen3vl_8b_bf16.safetensors

With ComfyUI updated, all required Qwen Image 2.1 model files downloaded, and the folder structure verified, you're ready to load the workflow and start upscaling images with native alpha channel support.

3. Running the Qwen Image 2.1 Upscale Workflow

Now that ComfyUI has been updated and all the required Qwen Image 2.1 model files are in place, it's time to download and load the Qwen Image 2.1 upscaler workflow into ComfyUI. The workflow includes the necessary nodes and settings to upscale images.

Step 1: Download the Workflow

First, download the Qwen Image 2.1 upscaler workflow JSON file. This file contains the complete workflow configuration and will allow you to quickly set up the upscaling process without having to build the workflow manually.

👉 Download Qwen Image 2.1 Upscaler Workflow JSON

Step 2: Load the Workflow

Once you have downloaded the workflow JSON file, open ComfyUI.

To load the workflow, simply drag and drop the JSON file into the ComfyUI interface. ComfyUI will automatically import the workflow and display all of the nodes and connections, including the image loader, diffusion model nodes, and alpha channel handling nodes.

Step 3: Input Your Image

Load the image you want to upscale into the Load Image node.

Step 4: Configure Options

Adjust the upscale factor (recommended 2× or 4×). Use the simple prompt inside the TextEncodeQwenImage21 node: “Keep colors exactly the same, upscale the image.”

Step 5: Run the Workflow

Click Run. The workflow processes a single input image and produces the upscaled result, allowing easy before-and-after comparison.

4. Examples Qwen Image 2.1 Upscale & Edit

Below are practical examples of the Qwen Image 2.1 upscaler workflow running in ComfyUI.

Example 1: Upscale + Background Removal

Load your image, adjust the upscale factor, and use the following prompt in the TextEncodeQwenImage21 node:

“Remove the background and make it fully transparent with an alpha channel. Keep the subject unchanged”

This produces a clean upscaled result with full transparency.

Example 2: Upscale + Creative Edit

Load your image, adjust the upscale factor, and use the following prompt in the TextEncodeQwenImage21 node:

“Keep colors exactly the same, upscale the image and change the plant in the background to a plume, and she has giving a wink, her hand resting on her chest”

This delivers an upscaled image with targeted creative edits applied during the process.

5. Conclusion

You now have a working ComfyUI setup for upscaling images with Qwen Image 2.1 while taking advantage of its native alpha channel support for background removal. The workflow is straightforward once the models are in place.

Qwen Image 2.1 serves as a strong alternative to SeedVR2 when you need more than basic upscaling. While SeedVR2 focuses purely on high-quality image upscaling, Qwen Image 2.1 excels at handling transparency and performing light edits during the upscale process. If you're interested in SeedVR2 instead, check out our guide here:

👉 SeedVR2: High-Quality AI Image Upscaling in ComfyUI

Results are impressive for both upscaling and transparency tasks, though very large images may require additional VRAM. Experiment with different upscale factors to find the best balance for your use case.