Qwen-Image

Qwen-Image

Turns written prompts into images and renders readable text inside them, in English and Chinese. Suits designers, artists and developers who work with the diffusers library.

Made by Qwen on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: Qwen-Image is a text to image model in the Qwen series, with 20.4B parameters. It generates pictures from written prompts and is built to place readable text inside them, especially Chinese. The card also describes image editing, and lists English and Chinese as its languages.

Who it suits: Designers, artists and storytellers who need pictures with legible words fit best. Developers who want to call an image model from Python with the diffusers library can use it too.

Industry | Who uses it | How they use it | Example
Design | Graphic designers | Make posters and signs that carry readable text | A shop sign with a Chinese neon light
Storytelling | Artists and writers | Create scenes in styles from photorealistic to anime | An impressionist painting from a short prompt
Software | Python developers | Call the model through the diffusers library | Generate a 16:9 image and save it as a file

If disk space is tight, you can skip it, since the full weights are large.

Where to use it: Use it from Python on a computer that can hold the weights.

• Python scripts that use the diffusers library
• Windows, macOS or Linux, where you download the files with Python
• A CUDA GPU when one is found, otherwise the CPU

Use case | Who | What they get back
Text-heavy images | Designers | Posters, signs and labels with readable English or Chinese text
Art in many styles | Artists and storytellers | Scenes from photorealistic to impressionist, anime or minimalist
Image edits | Everyday users | Style changes, adding or removing objects, text edits and pose changes, as the card describes them

How to set it up: You need Python and a lot of free disk space.

• Python on Windows, macOS or Linux
• Room for the weights, about 40.9 GB in BF16 or about 10.2 GB at 4-bit as an estimate
• The latest version of diffusers, which the card installs from its code repository
• A CUDA GPU if you have one, because the example falls back to the CPU otherwise

1. Install the latest version of diffusers, following the card.
2. Install the Hugging Face hub package.
3. Download the model files.

pip install -U huggingface_hub
hf download Qwen/Qwen-Image

How to use it: Load the pipeline in Python. The example picks bfloat16 on a CUDA GPU and float32 on the CPU.

from diffusers import DiffusionPipeline
import torch

model_name = "Qwen/Qwen-Image"

# Load the pipeline
if torch.cuda.is_available():
    torch_dtype = torch.bfloat16
    device = "cuda"
else:
    torch_dtype = torch.float32
    device = "cpu"

pipe = DiffusionPipeline.from_pretrained(model_name, torch_dtype=torch_dtype)
pipe = pipe.to(device)

Next, set a prompt variable to your own text. The script adds a short quality phrase for English or Chinese prompts, picks a size from a list of aspect ratios, runs 50 steps and saves the image as example.png.

positive_magic = {
    "en": ", Ultra HD, 4K, cinematic composition.", # for english prompt
    "zh": ", 超清，4K，电影级构图." # for chinese prompt
}

negative_prompt = " " # using an empty string if you do not have specific concept to remove

aspect_ratios = {
    "1:1": (1328, 1328),
    "16:9": (1664, 928),
    "9:16": (928, 1664),
    "4:3": (1472, 1140),
    "3:4": (1140, 1472),
    "3:2": (1584, 1056),
    "2:3": (1056, 1584),
}

width, height = aspect_ratios["16:9"]

image = pipe(
    prompt=prompt + positive_magic["en"],
    negative_prompt=negative_prompt,
    width=width,
    height=height,
    num_inference_steps=50,
    true_cfg_scale=4.0,
    generator=torch.Generator(device="cuda").manual_seed(42)
).images[0]

image.save("example.png")

Example: You write "A coffee shop entrance with a chalkboard price sign and a neon sign in Chinese", keep the 16:9 setting and run the script. It saves one image named example.png, 1664 pixels wide and 928 tall, built from that prompt.

For newcomers and experts: Qwen-Image turns a written description into a picture, and it is made to put readable words in that picture. A good first try is to download the files with the two Hugging Face commands. Then run the pipeline with a short prompt that asks for a sign with your own words on it.

Under the hood it is a 20.4B parameter model loaded through the diffusers DiffusionPipeline class. The example runs 50 inference steps with a true_cfg_scale of 4.0 and an empty negative prompt. The trade-off is size, since the BF16 weights are about 40.9 GB and the CPU path uses float32.

Write a prompt in English or Chinese
Add the quality phrase to the prompt
Choose an aspect ratio and size
Run 50 inference steps in the pipeline
Save the image as a file

Good to know: Other people have published 44 quantized and 108 fine-tuned versions of this model, which can help when the full weights are too big.

• License: Apache 2.0
• Cost to run: none named. The weights take about 40.9 GB in BF16 or about 10.2 GB at 4-bit, as an estimate
• Data it reaches: the model files from Hugging Face, your prompt and the image the script saves in the folder you run it from
• Not for: the source names no exclusions

On Hugging Face, 13 November 2026

• Task: text to image
• Library: diffusers
• Parameters: 20.4B
• Weights: about 40.9 GB in BF16; about 10.2 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 348,570
• Likes: 2,696
• License: apache-2.0
• Languages: en, zh
• Quantized versions: 44
• Fine-tunes: 108
• Tags: diffusers, safetensors, text-to-image

Links

• Model: https://huggingface.co/Qwen/Qwen-Image
• Files: https://huggingface.co/Qwen/Qwen-Image/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download Qwen/Qwen-Image

Then follow the usage in its model card. Nothing in this kit runs by itself.