Download Qwen Image 2.1

Official Qwen Image 2.1 download sources

The model card links four official locations. Use these rather than re-uploads, so the files and the license text match what the Qwen team published.

ComfyUI users normally download a repackaged set instead: Comfy-Org/Qwen-Image-2.1 on Hugging Face holds single-file versions of the diffusion model, text encoder and VAE that match the built-in workflow templates.

What is in the download

The official repository is split into the three parts the pipeline loads. Sizes below were read from the Hugging Face file listing on 4 October 2026.

If disk space or VRAM is tight, the ComfyUI repack offers an INT8 diffusion model of about 7.3 GB next to the 14.2 GB BF16 file, and community GGUF quantizations go down to roughly 2.3 GiB.

Download commands

pip install -U huggingface_hub
hf download Qwen/Qwen-Image-2.1 --local-dir ./Qwen-Image-2.1

pip install -U modelscope
modelscope download --model Qwen/Qwen-Image-2.1 --local_dir ./Qwen-Image-2.1

Older installs of huggingface_hub use the command name huggingface-cli download with the same arguments. Calling from_pretrained in Diffusers also downloads the files automatically into the Hugging Face cache on first use.

Run Qwen Image 2.1 locally with Diffusers

The model card lists torch 2.4.0 or newer, transformers 5.17 or newer, and Diffusers installed from the GitHub main branch, plus accelerate and pillow. The published example runs 40 inference steps at 2048 x 2048.

pip install "torch>=2.4.0" "transformers>=5.17" accelerate pillow
pip install git+https://github.com/huggingface/diffusers

import torch
from diffusers import QwenImage21Pipeline

pipe = QwenImage21Pipeline.from_pretrained(
    "Qwen/Qwen-Image-2.1", torch_dtype=torch.bfloat16
)
pipe.enable_model_cpu_offload()  # or pipe.to("cuda") with enough VRAM

image = pipe(
    prompt="A botanical poster with the headline \"GROW TOGETHER\"",
    width=2048, height=2048,
    num_inference_steps=40,
).images[0]
image.save("qwen-image-2-1.png")

The same pipeline accepts an image argument for editing and the RGBA prompt pattern for transparent output. The model card does not publish a minimum VRAM figure; enable_model_cpu_offload is its documented memory option.

Read the license before you download

Qwen-Image-2.1 is released under the Qwen Research License Agreement, not Apache 2.0. The agreement grants use, modification and redistribution for non-commercial purposes only, and defines non-commercial as research or evaluation. Commercial use requires a separate license from the Qwen team.

This is a summary, not legal advice. Read the LICENSE file in the repository; it is the binding text.

No GPU? Try Qwen Image 2.1 online

A 33 GB download and a recent GPU are a lot to commit before you know whether the model suits your work. In this studio, Qwen Image 2.1 runs through a hosted API: write a prompt, pick an aspect ratio, and get one image back. Generation costs 1 credit per image, and nothing is installed on your device.

qwenimage21.pro is an independently operated workspace. It is not affiliated with or endorsed by Alibaba or the Qwen team, and it does not host or redistribute model weights.